Compare text and language model APIs
Pick a model. Call it. Switch any time.
Explore language models for chat, reasoning, coding, extraction, agents, and structured generation. Every model uses the OneInfer API surface, so teams can compare and switch without rebuilding their integration.
163 text models are listed today. All of them use one OneInfer API key and pay-as-you-go wallet, so switching models means changing one model id.
Choose and call a text model in 4 steps
The list ranks them by the average of their input and output price per million tokens. Different billing units are never mixed, so the order is fair.
Lowest comparable price: Meta: Llama 3.1 8B Instruct
Currently the lowest-priced option in the average of available input and output prices per one million tokens group. Different billing units are never mixed.
$0.033 avg / 1M tokens
View pricing and providersAvailable text models
Comparable models are ordered by their current normalized price. Pricing content refreshed November 9, 2026.
163 models
| Model |
|---|
 Meta: Llama 3.1 8B InstructLowest pricedeepinfra Llama-3.1-8B-Instruct is Meta's optimized 8B parameter instruction-tuned model featuring enhanced reasoning, safety alignment, and tool-calling capabilities. Represents a significant improvement over original Llama 3 8B with better instruction following and reduced hallucination.
|
deepinfra Lunaris is Sao10K's Llama 3 8B merge for creative writing and roleplay. It has an 8K-token context window.
|
 thinkingmachines GPT-OSS-20B is OpenAI's efficiency-optimized 20-billion-parameter transformer featuring architectural innovations from GPT-4. Designed for accessible deployment, it delivers 85% of GPT-OSS-120B's capability at 6× lower resource requirements. Includes sparse attention mechanisms, constitutional AI safeguards, and deterministic output options.
|
 deepinfra Mistral Small 3 is Mistral AI's 24B model tuned for low latency across everyday tasks. It has a 32K-token context window.
|
 novita Gemma 3 12B Instruct is a 12-billion-parameter multimodal model from Google, built from the same research as the Gemini models. It is instruction-tuned for dialogue and excels at text generation and image understanding tasks like question answering, summarization, and reasoning.
|
deepinfra Gemma 3 4B is a small Gemma 3 open model from Google that reads text and images, for fast, low-cost tasks. It has a 128K-token context window.
|
deepinfra Mistral-Nemo-Instruct-2407 is a 12-billion-parameter instruction-tuned language model developed jointly by Mistral AI and NVIDIA. It features a 128k token context window, strong multilingual capabilities, and excels at tasks like text generation, code generation, and reasoning. It uses the efficient Tekken tokenizer and is released under the Apache 2.0 license.
|
cerebras GPT-OSS-120B is OpenAI's open-source 120-billion-parameter transformer model featuring architectural innovations from GPT-4. Optimized for large-scale deployment with custom FP8 quantization, it delivers state-of-the-art reasoning capabilities while maintaining 3× better throughput than comparable models. Includes constitutional AI safeguards and deterministic output options.
|
deepinfra Phi-4 is Microsoft's 14B small language model, trained on high-quality synthetic and curated data for strong reasoning at a low cost. It has a 16K-token context window.
|
 novita Ling 3.0 Flash is InclusionAI's highly efficient open-weight hybrid reasoning model designed for reasoning, coding, long-context understanding, and complex agentic workflows. It uses a sparse Mixture-of-Experts architecture with 124B total parameters and only 5.1B activated parameters per token. The model combines Kimi Delta Attention and gated Multi-Head Latent Attention in a 5:1 hybrid-linear architecture, with 512 routed experts and 8 experts activated per token. Ling 3.0 Flash supports native hybrid reasoning, allowing thinking to be enabled or disabled, and is trained for end-to-end agentic execution across coding, general productivity, and deep-research environments. Its architecture is optimized for high inference efficiency and long-context workloads while delivering performance competitive with substantially larger models.
|
 deepinfra Nemotron 3.5 Lightning is NVIDIA's fastest open model, for always-on agents and high-volume tasks. It reasons, calls tools and has a 256K-token context window.
|
novita NVIDIA Nemotron-3 Nano 30B A3B is NVIDIA's open reasoning model featuring a hybrid Mamba-Transformer Mixture-of-Experts (MoE) architecture. It consists of 23 Mamba-2 and MoE layers, plus 6 Attention layers. The model is built for building specialized AI agents, chatbots, and RAG systems, providing high compute efficiency. It natively supports configurable reasoning depth through a 'thinking budget' and acts as a general-purpose reasoning and chat model intended for English, coding languages, and several other supported languages.
|
 oneinfer Solar Mini 4 is Upstage's lightweight Mixture-of-Experts language model built for fast, economical agentic workloads. With 35B total parameters and only 3B activated per token, it delivers efficient inference while supporting a 512K-token context window for large documents and long-running workflows. The model combines fluent Korean understanding with strong English and Japanese capabilities and supports reasoning, tool calling, structured outputs, and long-context processing. Solar Mini 4 is particularly suited to high-volume agents, routing, classification, document analysis, and applications where latency and serving cost are important.
|
deepinfra Gemma 3 27B is Google's largest Gemma 3 open model. It reads text and images and answers in text, with a 128K-token context window and tool calling.
|
 deepinfra Qwen3.5 9B is a small Qwen3.5 model from Alibaba that reads text and images, reasons and calls tools. It has a 256K-token context window.
|
novita Ling-3.0-flash-Fin is InclusionAI's finance-enhanced reasoning and agentic model built on Ling-3.0-flash through continued training on high-quality financial data. It combines 124B total parameters with only 5.1B activated parameters per token and a 256K context window, providing efficient inference for long-horizon financial workflows. The model is designed for end-to-end financial research, source-grounded financial search, multi-document reasoning across annual reports, earnings releases and regulatory filings, valuation and financial modeling, spreadsheet analysis, scenario analysis, and professional report preparation. It supports tool-intensive workflows and native thinking, making it suitable for investment research, financial analysis, banking workflows, financial agents, and other complex finance-focused applications.
|
novita Ling 3.0 Flash VL is InclusionAI's native multimodal reasoning and agentic model built on Ling 3.0 Flash. It extends the language, reasoning, and long-context capabilities of the base model with native image and video understanding, bringing visual information into the complete process of understanding, reasoning, acting, and verification. The model contains 124B total parameters while activating only 5.5B parameters per token and supports up to a 256K-token context window. It is designed for complex visual understanding, document and chart analysis, multimodal reasoning, visual verification, web and software interface understanding, computer-use-style workflows, coding, and tool-driven agentic tasks. Thinking is enabled by default and can be disabled per request.
|
deepinfra Mistral Small 3.2 is Mistral AI's 24B model with better instruction following and tool calling than 3.1. It reads text and images and has a 128K-token context window.
|
deepinfra Granite 4.2 8B is the mid-size reasoning model in IBM's open Granite 4.2 family, for reasoning, tool calling and enterprise tasks. It has a 128K-token context window.
|
groq gpt-oss-safeguard-20b is OpenAI's open-weight safety reasoning model, fine-tuned from gpt-oss-20b to classify content against a policy you write, and explain its decision. It has a 128K-token context window.
|
deepinfra Llama Guard 4 is Meta's 12B safety classifier. It labels a conversation's text and images as safe or unsafe, and names the categories it breaks.
|
deepinfra Qwen3 14B is a dense Qwen3 model from Alibaba that switches between a thinking mode for reasoning and a fast mode. It calls tools and has a 40K-token context window.
|
deepinfra Qwen3 32B is the largest dense Qwen3 model from Alibaba, with a thinking mode for reasoning and a fast mode. It calls tools and has a 40K-token context window.
|
deepinfra Llama 4 Scout is Meta's natively multimodal mixture-of-experts model, with 17B active parameters across 16 experts. It reads text and images, calls tools, and has a 320K-token context window.
|
deepinfra Nemotron Content Safety 3.5 is NVIDIA's compact safety classifier for text and images. It flags unsafe content in prompts and responses by category.
|
openai GPT-5 Nano is a lightweight, low-latency language model designed for cost-efficient text generation, chat, and tool-calling workloads.
|
 novita GLM-4.7-Flash is a 30B-A3B Mixture-of-Experts (MoE) model designed for lightweight deployment that balances performance and efficiency. It is positioned as the strongest model in the 30B class, with strong performance on reasoning and agentic benchmarks.
|
 bytedance Seed 2.0 Mini is ByteDance's fast, low-cost Seed 2.0 model with a 256K context and image input, and the successor to Seed 1.6 Flash.
|
novita Gemma 4 26B A4B is an efficient, open-weights Mixture-of-Experts (MoE) multimodal model from Google DeepMind. With only 3.8B active parameters out of 25.2B total, it runs nearly as fast as a 4B model while delivering performance close to the 31B dense model. Features a 256K token context window, supports interleaved text and image inputs, and excels at reasoning, coding, and agentic workflows.
|
deepinfra Nemotron 3 Super is NVIDIA's hybrid mixture-of-experts reasoning model, with 12B of its 120B parameters active per token. It calls tools and has a 256K-token context window.
|
novita Llama-3.3-70B-Instruct is Meta's flagship 70B parameter model featuring enhanced reasoning, 128K context, and enterprise-grade instruction following. Represents a significant evolution over Llama 3.1 with improved tool integration, safety alignment, and complex task handling.
|
novita Gemma 4 31B is a state-of-the-art, open-weights multimodal dense model from Google DeepMind. It features a 256K token context window, excels at reasoning, coding, and agentic workflows, and supports interleaved text and image inputs.
|
openai GPT-6 Luna is OpenAI's most efficient GPT-6 model for focused, high-volume workloads. It combines GPT-6 reasoning and agentic capabilities with substantially lower serving cost, making it suitable for scalable automation, coding assistance, information extraction, classification, structured generation, multimodal understanding, and repeatable agent workflows. The model supports a 1.05M-token context window, up to 128K output tokens, configurable reasoning effort, image understanding, function calling, structured outputs, web search, file search, and computer use.
|
novita Qwen3.8-Flash is Qwen's high-speed multimodal reasoning model designed for coding, agentic workflows, visual understanding, long-context processing, and high-concurrency applications. It natively supports a 1M-token context window and can process text, images, video, lengthy documents, and large codebases. The model combines strong reasoning and generation capabilities with efficient inference, and supports built-in tools for agentic and developer workflows.
|
zai GLM-5.3-Flash is Z.AI's natively multimodal Mixture-of-Experts model in the GLM-5 family, designed for high-performance coding, agentic engineering, multimodal reasoning, long-context processing, and efficient inference. It has 320B total parameters with 18B active parameters and introduces a hybrid architecture combining sparse and linear attention together with Manifold-Constrained Hyper-Connections for improved scaling and long-context efficiency. The model supports controllable reasoning effort and is optimized to deliver frontier-level capability at Flash-tier serving cost.
|
deepinfra NVIDIA Nemotron 3.5 Super VL is an open-weight vision-language model released in October 2026: a hybrid mixture-of-experts with 120B total and 12B active parameters. It reads text, images and video, supports reasoning and tool calling, and has a 1M-token context window.
|
 deepinfra DeepSeek-V3.2 is a general-purpose Mixture-of-Experts (MoE) large language model from DeepSeek AI that harmonizes high computational efficiency with superior reasoning and agent performance. It features 685 billion total parameters, supports a 164Ktoken context window, and is natively calibrated for tool use and function calling in complex agentic workflows.
|
deepinfra Qwen3-235B-A22B-Instruct-2507 is Alibaba Cloud's frontier mixture-of-experts model featuring 235B total parameters with 22 specialized experts and 4 active per token. Designed for superhuman reasoning and multilingual mastery, it achieves state-of-the-art performance across technical, creative, and analytical domains with 256K context handling. Optimized for distributed inference across H100 GPU clusters.
|
deepinfra Qwen3 30B A3B is a Qwen3 mixture-of-experts model from Alibaba with 3B active parameters, with a thinking mode for reasoning and a fast mode. It has a 40K-token context window.
|
novita Hy3 is Tencent's 295B mixture-of-experts model, with 21B active parameters, for reasoning, coding and agentic tasks, with three reasoning modes. It calls tools and has a 256K-token context window.
|
openai GPT-4o Mini is a distilled 400B-parameter version of GPT-4o optimized for cost-efficient omnimodal applications. Maintains core multimodal capabilities with 70% of GPT-4o's performance at 40% of the computational cost, featuring real-time audio/image processing optimizations.
|
oneinfer Solar Pro 4 is Upstage's flagship language model built for complex agentic workflows and long-running professional tasks. It is designed to reason across large documents, execute multi-step tool-driven workflows, work with terminal environments, and produce reliable business deliverables across extended sessions. With a 524K-token context window and up to 128K output tokens, Solar Pro 4 can process multiple reports, contracts, datasets, and other large inputs within a single workflow. It supports configurable reasoning effort, function calling, structured outputs, and strong English, Korean, and Japanese capabilities, making it suitable for enterprise agents, coding, document analysis, research, information extraction, and complex workflow automation.
|
 minimax MiniMax-M3.1-Flash is MiniMax's latest M-series model for agentic reasoning, tool use, coding and long-context work. It always thinks before answering, with adjustable depth, reads text and images and has a 1M-token context window.
|
 sarvam Sarvam-105B is an advanced Mixture-of-Experts (MoE) model with 10.3B active parameters, designed for superior performance across a wide range of complex tasks. It is highly optimized for complex reasoning, with particular strength in agentic tasks, mathematics, and coding.
|
deepinfra Qwen-2.5-72B-Instruct is Alibaba's flagship 72B parameter model featuring enhanced reasoning, 128K context, and advanced tool-calling capabilities. Excels in multilingual tasks, technical domains, and complex instruction following with improved safety alignment.
|
deepinfra Qwen3 VL 30B A3B Instruct is a mixture-of-experts Qwen3 vision-language model with 3B active parameters, for images, documents and screens. It has a 256K-token context window.
|
novita Qwen3.6-35B-A3B is a Mixture-of-Experts (MoE) vision-language model with 35B total and 3B activated parameters. It is built for agentic coding and repository-level reasoning, and features a 262K context length with thinking preservation capabilities.
|
deepinfra Qwen3 VL 235B A22B Instruct is Alibaba's largest Qwen3 vision-language model, for understanding images, documents and screens. It calls tools and has a 256K-token context window.
|
zai GLM-4.6V is a vision-language model designed for cloud and high-performance clusters. It introduces native multimodal function calling, interleaved image-text generation, and advanced document understanding. It supports a 128K context window and achieves SoTA visual understanding among models of similar scale.
|
novita Ling 3.1 Flash is InclusionAI's large-scale hybrid reasoning Mixture-of-Experts model designed for coding, agentic workflows, professional knowledge work, long-context analysis, finance, healthcare, and complex multi-step tasks. It contains approximately 560 billion total parameters while activating about 25 billion parameters per token, enabling high model capacity with more efficient inference. The model supports hybrid reasoning, tool calling, long documents, large codebases, extended task histories, and multi-step agent execution. It is optimized for software engineering, terminal-based agents, workflow automation, financial analysis, research, healthcare reasoning, and other tool-intensive professional applications.
|
bytedance DeepSeek V4.1 Flash Economy runs the same weights as DeepSeek V4.1 Flash at a lower price, generating slightly more slowly. Suited to high-volume work where cost matters more than speed.
|
novita WizardLM-2 8x22B is a 141-billion-parameter Mixture of Experts (MoE) model developed by Microsoft AI. It is the flagship model of the WizardLM-2 family, designed for state-of-the-art performance in complex chat, reasoning, multilingual tasks, and agent-like capabilities. It is built upon the Mixtral-8x22B architecture and trained using a fully AI-powered synthetic training system .
|
deepinfra DeepSeek-V3.1 is a 1.3T-parameter Mixture-of-Experts model with 236B active parameters per token. It features enhanced reasoning capabilities, extended 1M token context window, and improved multilingual support with advanced tool usage and code generation capabilities.
|
zai GLM-4.5-Air is the lighter GLM-4.5 model from Z.ai, a 106B mixture-of-experts with 12B active parameters, for agentic and coding tasks at a lower price. It has thinking and non-thinking modes and a 128K-token context window.
|
deepinfra DeepSeek-V3 is a 7B-parameter efficient model optimized for high-speed inference. Delivers fast response times with strong reasoning capabilities, supporting 128K context and tool calling at consumer-grade hardware requirements.
|
novita Step 3.7 Flash is StepFun's multimodal mixture-of-experts reasoning model (198B total, about 11B active) for fast agentic workflows. It reads text and images, calls tools and has a 256K-token context window.
|
openai GPT-5.6 Luna is the fastest, most cost-efficient model in the GPT-5.6 family. It is designed for high-volume, latency-sensitive workloads such as classification, data extraction, request routing, and draft generation. It shares the same 1.05M context window as its siblings
|
openai GPT-5.4 Nano is the smallest and fastest model in the GPT-5.4 lineup, designed for ultra-low latency and low-cost API usage at high throughput. It is optimized for short-turn tasks like classification, extraction, ranking, and lightweight sub-agent work.
|
oneinfer MiniMax M3 is an open model with sparse block-level attention designed for coding, agentic workflows, and multimodal chat. It supports a 1M token context window and operates natively across text, image, and video.
|
novita MiniMax-M2.7 is a next-generation large language model designed for autonomous, real-world productivity and continuous improvement. Built to actively participate in its own evolution, M2.7 integrates advanced agentic capabilities through multi-agent collaboration, enabling it to plan, execute, and refine complex tasks across dynamic environments.
|
novita MiniMax M2.5 is an open-source MoE model with 230B total parameters (10B active) and a 200K token context window, released February 2026 under a Modified MIT license. SOTA in coding, agentic tool use, search, and office work — scoring 80.2% on SWE-Bench Verified, 51.3% on Multi-SWE-Bench, and 76.3% on BrowseComp. Trained across 200K+ real-world RL environments in 10+ programming languages.Matches Claude Opus 4.6 speed while completing SWE-Bench 37% faster than M2.1.
|
deepseek DeepSeek-V4-Flash is a highly efficient Mixture-of-Experts (MoE) language model featuring 158B total parameters with only 13B activated during inference. It utilizes a Hybrid Attention Architecture (combining Compressed Sparse Attention and Heavily Compressed Attention) to natively and efficiently support a 1-million-token context window. The model offers switchable reasoning modes (Non-think, Think High, and Think Max) and excels at high-speed generation for reasoning, coding, tool-calling, and agentic tasks.
|
 thinkingmachines Inkling-Small is a general-purpose multimodal Mixture-of-Experts (MoE) model featuring 266B total parameters with only 12B active during inference. Natively processing text, image, and audio inputs, it supports a 256K token context window and variable thinking effort. Despite being a quarter of the size of the flagship Inkling model, it matches or exceeds its predecessor on key coding and agentic tasks while requiring significantly less compute.
|
deepinfra DeepSeek-V4.1-Flash is DeepSeek's multimodal Mixture-of-Experts model designed for advanced reasoning, coding, long-horizon agentic workflows, visual understanding, cybersecurity, and long-context processing. It has 552B backbone parameters while activating only 8B parameters per token during prefill and 16B during decoding. The model uses a 40-layer Causal Encoder-Decoder architecture together with Compressed Sparse Attention 2, Single-Pass mHC, Engram conditional memory, and DSpark speculative decoding to reduce inference and KV-cache costs. It natively processes text and images, supports a 1M-token context window, tool calling, and continuously configurable reasoning effort from 1 to 100.
|
together_ai Beam 1.0 is Reflection's open-weight reasoning model for coding and long-horizon agentic work, a sparse mixture-of-experts with 501B total and 23B active parameters. It calls tools and has a 1M-token context window.
|
zai GLM-5.3-FlashX is Z.AI's high-speed serving variant of GLM-5.3-Flash, designed for responsive coding, multimodal understanding, long-horizon agentic workflows, tool use, and long-context processing. It is built on the GLM-5.3-Flash foundation with 320B total parameters and 18B active parameters, using a hybrid architecture that combines sparse and linear attention for efficient long-context inference. FlashX supports a 1M-token context window, reasoning, function calling, and native multimodal understanding while prioritizing accelerated inference with generation speeds of up to 200 tokens per second.
|
deepinfra Qwen3-Next-80B-A3B-Instruct is an 80B-parameter MoE model with 3B active parameters per token, optimized for instruction following and complex reasoning tasks. Features enhanced multilingual capabilities, advanced tool use, and extended context handling with improved efficiency through expert routing.
|
together_ai Muse Glimmer 30B is a dense, open-weight multimodal model from Meta Superintelligence Labs, distilled from Muse Spark and optimized for local autonomous agentic workflows on consumer hardware. It combines multi-step reasoning, reliable schema-based tool calling, and failure recovery with multimodal understanding via a built-in perception encoder. It supports interleaved text and image input, controllable reasoning strength, and is highly compatible with agentic orchestration frameworks.
|
deepinfra Llama-Guard-4-12B is a 12B-parameter safety classifier based on Llama 3 architecture, specializing in content moderation and policy enforcement. It performs multi-category risk assessment for AI-generated content with enterprise-grade policy customization capabilities and low false positive rates.
|
deepinfra Qwen3-Coder-480B-A35B-Instruct is a massive 480B-parameter MoE model specialized in code generation and technical problem-solving. Featuring 35B active parameters per token through expert routing, it delivers state-of-the-art coding assistance with FP8 quantization for efficient deployment.
|
deepinfra DeepSeek V4 Flash Vision is DeepSeek's experimental V4 Flash model that also reads images. It reasons, calls tools and has a 1M-token context window.
|
openai GPT-4.1 Mini is a distilled 400B-parameter version of GPT-4.1 optimized for cost-efficient deployment. Features the same 256-expert MoE architecture with selective expert activation, delivering 80% of GPT-4.1's capability at 30% of the inference cost.
|
cerebras Qwen3.8-27B is a compact, deployment-friendly dense multimodal model from Alibaba's Qwen3.8 family. Natively supporting text, image, and video inputs, it handles tasks ranging from STEM diagrams and document understanding to long-form video analysis. It features flexible thinking control with adjustable reasoning effort and is designed for autonomous planning, coding, professional work, multimodal reasoning, and long-horizon agentic workflows.
|
novita Qwen3.5-35B-A3B is a Mixture-of-Experts (MoE) vision-language model with 35B total parameters and 3B activated. It balances efficiency and performance with Gated Delta Networks and sparse MoE, supporting a 262K context length for agentic coding and reasoning tasks.
|
openai OpenAI's most compact and cost-effective GPT-5 model, released August 7, 2025. Designed for high-volume, latency-sensitive applications where the full capabilities of larger models are not required, while maintaining strong performance on common tasks.
|
bytedance Seed 2.0 Lite is ByteDance's balanced Seed 2.0 model with a 256K context and image input, and the successor to Seed 1.6 and Seed 1.8.
|
zai GLM-4.5V is Z.ai's vision-language model built on GLM-4.5-Air (106B, 12B active). It reasons over images, documents and screenshots, with a thinking mode that can be switched off.
|
novita Qwen3.5-27B is an open-weight vision-language model with a hybrid architecture combining Gated Delta Networks and Gated Attention. It features a 262K native context length, unified vision-language foundation, and supports scalable reinforcement learning for robust agentic tasks
|
zai GLM-4.6 is a large language model with improvements over GLM-4.5, featuring a 200K context window, superior coding performance, advanced reasoning, and stronger agentic capabilities in tool use and search-based agents. It also offers refined writing and more natural role-playing.
|
novita MiniMax-M1 is MiniMax's open-weight hybrid-attention reasoning model, a 456B mixture-of-experts with 45.9B active parameters. This is its 80K thinking-budget version, with a 1M-token context window.
|
oneinfer Mistral Large 4 is Mistral AI's flagship mixture-of-experts model, with 1.05T total and 52B active parameters, in public preview since October 2026. It reads text and images, supports adjustable reasoning and tool calling, and has a 1M-token context window.
|
zai GLM-4.7 is a large language model focused on coding, agentic tasks, and complex reasoning. It introduces Interleaved Thinking, Preserved Thinking, and Turn-level Thinking for improved stability and performance in multi-turn, long-horizon tasks. It shows significant gains over its predecessor in multilingual agentic coding, terminal-based tasks, and tool use.
|
zai GLM-4.5 is Z.ai's 355B mixture-of-experts model, with 32B active parameters, built for agents: tool calling, coding and web browsing. It switches between a thinking mode and instant answers and has a 128K-token context window.
|
 novita Kimi-K2-Instruct is Moonshot AI's flagship 100-billion-parameter instruction-tuned model featuring a hybrid dense-MoE architecture with 200K context handling. Optimized for complex reasoning, multilingual dialogue, and long-form document understanding. Specializes in Chinese and English technical domains with enhanced safety alignment.
|
deepinfra DeepSeek-R1-0528 is an open-source 7B-parameter language model optimized for long-context reasoning, multilingual understanding, and coding tasks. Features efficient deployment and strong reasoning capabilities.
|
bytedance Dola-Seed 2.1 Turbo is a next-generation multimodal large model by ByteDance designed for real-world productivity, high-value production tasks in enterprise R&D, and large-scale agentic scenarios. It features comprehensive upgrades in coding engineering delivery, long-horizon agent task execution, and multimodal understanding, along with stronger autonomous planning and dynamic self-repair capabilities.
|
 anthropic Claude Haiku 5.5 is Anthropic's fastest, cheapest, and most capable Haiku-class model, designed for high-volume, latency-sensitive, and cost-sensitive workloads. It is optimized for classification, extraction, routing, summarization, database queries, live customer support, browser use, subagent workflows, and lightweight coding tasks. The model supports a 1M-token context window, up to 128K output tokens, adaptive thinking with configurable effort, vision, tool use, computer use, and agentic workflows. Haiku 5.5 is particularly suited to applications requiring high throughput and low latency while retaining strong reasoning, coding, computer-use, and knowledge-work capabilities.
|
 grok Grok Build 0.1 is xAI's model for coding agents, built for agentic software engineering and tool calling. It has a 256K-token context window.
|
novita Kimi-K2-0905 is a 72B-parameter multimodal language model optimized for long-context understanding and complex reasoning. Features enhanced multilingual capabilities, advanced tool usage, and strong performance in mathematical and logical reasoning tasks with extended context handling.
|
novita Kimi K2 Thinking is Moonshot AI's agentic reasoning model in the Kimi K2 family. It thinks step by step between tool calls and has a 256K-token context window.
|
bytedance Dola-Seed 2.0 Pro is a deep-thinking multimodal model from ByteDance offering strong reasoning capabilities. It is optimized to support complex multi-step agentic workflows and advanced vision-language processing tasks.
|
bytedance Dola-Seed 2.0 Code is ByteDance's dedicated coding model optimized for enterprise-level programming needs. It excels in agentic task planning with rational decomposition and dynamic self-repair, while inheriting the Seed 2.0 family's robust multimodal perception to read charts, screenshots, and diagrams for contextually accurate code generation.
|
novita Kimi K2.5 is a large open-weight MoE model with 1.1 trillion parameters, known for strong vision capabilities via MoonViT-3D and agent swarm orchestration.
|
novita Qwen3.6-27B is Alibaba's multimodal hybrid-thinking model prioritizing stability and real-world utility. It delivers top performance for its size, supports a 256K context window across 201 languages, and excels in agentic coding, vision, and chat tasks. It features a new option to preserve reasoning context from historical messages.
|
novita Qwen3.5-122B-A10B is a mid-tier multimodal vision-language MoE model designed for native multimodal agent applications. It utilizes a hybrid 'Gated DeltaNet' and sparse MoE architecture to process text, image, and video inputs with early fusion, balancing high performance with computational efficiency.
|
deepinfra Hy4 Preview is Tencent's 770B mixture-of-experts flagship, with 49B active parameters, for reasoning, coding and agentic tasks. It calls tools and has a 1M-token context window.
|
grok Grok 4.20 Reasoning is a high-performance model by xAI designed for complex logic and math, scientific and technical analysis, multi-step investigations, and high-stakes tasks where accuracy matters most. It incorporates a 'thinking' mechanism before responding and offers strict prompt adherence alongside agentic tool calling.
|
grok Grok 4.3 is a flagship reasoning model from xAI leveraging a four-agent deliberation system where internal sub-agents debate the best response. It features native video input and leverages real-time data streams from the X platform for current-events knowledge. It is optimized for long-document analysis, conversational research, and complex multi-step agentic workflows.
|
grok Grok 4.20 Non-Reasoning is xAI's Grok 4.20 without the thinking step: it answers straight away, for lower latency and cost than Grok 4.20. It has a 1M-token context window.
|
novita Kimi-K2.6 is an open-source native multimodal agentic model developed by Moonshot AI, supporting long-horizon coding capabilities and agentic task orchestration scaling to 300 sub-agents.
|
novita GLM-5 is a 754B-parameter Mixture-of-Experts model from Z.ai. It features DeepSeek Sparse Attention (DSA) for efficient long-context handling, leading open-source models in long-horizon planning and agentic tasks.
|
novita Qwen3.5-397B-A17B is Alibaba Cloud's largest open-weights multimodal foundation model. Unifying the previously separate text and vision lines, it features a hybrid Mixture-of-Experts architecture with early fusion vision-language training. It operates in thinking mode by default and delivers state-of-the-art performance across chat, reasoning, tool-calling, and complex agentic workflows.
|
deepinfra NVIDIA Nemotron-3 Ultra 550B A55B is a frontier-scale large language model from NVIDIA, designed for strong agentic, reasoning, and conversational capabilities. It employs a hybrid Latent Mixture-of-Experts (LatentMoE) architecture, utilizing interleaved Mamba-2, MoE, and Attention layers, alongside Multi-Token Prediction (MTP) for faster inference. With a 1-million-token context window and a configurable reasoning mode, it excels at complex autonomous agents, long-context analysis, and deep research workflows.
|
zai GLM-5.3 is Z.ai's advanced Mixture-of-Experts model, building upon the GLM-5.2 base architecture with significantly scaled post-training. It is designed specifically for complex coding, long-horizon tasks, and autonomous agent workflows. The model sets a new state-of-the-art for open models in agentic coding and demonstrates strong emergent cybersecurity capabilities, excelling in vulnerability discovery and exploitation chaining.
|
deepseek DeepSeek-V4-Pro is a flagship Mixture-of-Experts (MoE) language model with 862B total parameters (49B activated) and a 1-million-token context window. It features a Hybrid Attention Architecture combining Compressed Sparse Attention and Heavily Compressed Attention for efficient long-context processing. With multiple distinct reasoning modes (such as Think Max for maximum reasoning effort), it is built for advanced mathematical reasoning, software engineering, tool use scenarios, and complex long-horizon agentic workflows.
|
novita Kimi K2.7 Code is a coding-focused agentic model built upon Kimi K2.6. With substantial improvements on real-world long-horizon coding tasks, it strengthens end-to-end task completion across complex software engineering workflows while improving token efficiency, reducing thinking-token usage by approximately 30% compared with Kimi K2.6.
|
novita Qwen3.7-Max is a flagship model from Alibaba Cloud, representing the most capable version in the Qwen 3.7 series. It is designed for top-tier performance across reasoning, coding, and agentic tasks, and is typically accessed via the official Qwen API.
|
thinkingmachines Inkling is a flagship open-weights multimodal Mixture-of-Experts (MoE) model from Thinking Machines Lab featuring 975B total parameters with 41B active per token. It natively processes text, image, and audio inputs and supports a 256K context window. The model offers a controllable 'thinking-effort' dial to balance compute cost against reasoning quality, excelling across coding, agentic workflows, and general reasoning tasks.
|
openai GPT-5.4 Mini is a compact, cost-efficient model designed for reliable performance across high-volume, everyday AI workloads. It offers fast, efficient reasoning, strong multimodal capabilities, and approaches the performance of the larger GPT-5.4 model on coding and agentic tasks while running more than 2x faster than its predecessor.
|
meta Muse Spark 1.3 is Meta's natively multimodal reasoning model optimized for long-horizon agentic workflows, advanced coding, tool use, computer interaction, and complex multi-step tasks. It is designed to sustain multiple workflows within long conversations, generate and manage context across conflicting sources, identify and correct gaps in its plans, and efficiently complete real-world engineering and knowledge-work tasks. Compared with Muse Spark 1.2, it improves coding and agentic performance while using fewer tool calls and fewer tokens.
|
|
bytedance GLM-5.2 is Z.ai's latest flagship open-weights MoE model engineered specifically to dominate long-horizon autonomous coding and engineering tasks. It features a solid 1-million-token context window, multiple thinking effort levels, and an improved IndexShare architecture that significantly boosts reasoning and agentic performance.
|
openai GPT-6.1 Terra is OpenAI's fast, efficient GPT-6.1 model for visual reasoning, computer use and clear, natural writing. It reads text and images, reasons with adjustable effort, calls tools and has a 1.05M-token context window. Prompts over 272K tokens are billed at OpenAI's long-context rates.
|
anthropic Claude Haiku 4.5 is Anthropic's small, fast model from October 2025. It reads text and images, supports extended thinking, calls tools and has a 200K-token context window.
|
elevenlabs Eleven Realtime v1 Mini is ElevenLabs' live speech-to-speech model, the faster, lower-cost size, for everyday support and assistant calls. Connect to WS /v1/realtime, stream 8, 16 or 24 kHz PCM (or 8 kHz G.711) microphone audio in and get 24 kHz spoken replies back, using the OpenAI Realtime event format. Speaks 90+ languages with ElevenLabs voices and supports function tools. Billed per connected minute plus the LLM tokens each conversation uses.
|
novita Qwen3.8-Max is Alibaba's 2.4-trillion-parameter flagship multimodal large language model. Designed for long-horizon professional workflows, it excels in full-stack coding, agentic development, data analysis, and office automation. It processes text, images, video, and documents natively, and is positioned as a leading frontier model.
|
grok Grok 4.5 is xAI's frontier model built for coding, agentic tasks, and knowledge work. Trained on datasets spanning science, engineering, math, and code (in partnership with Cursor), it features a 1.5 trillion parameter architecture and exceeds comparable models in real-world software engineering tasks.
|
grok Grok 4.6 is an incremental flagship upgrade over Grok 4.5, utilizing the same 1.5 trillion parameter foundation but enhanced with superior supervised fine-tuning and reinforcement learning via the Grok Build harness. It brings frontier intelligence to long-running AI agents, complex multi-step tasks, and software development, offering aggressive verification and highly competitive cost-per-task efficiency.
|
novita Qwen3.8-2.4T-A95B is Qwen's largest open-weight Qwen3.8 reasoning model, built for advanced coding, professional work, research, and long-horizon agentic tasks. It uses a Mixture-of-Experts architecture with 2.4 trillion total parameters and 95 billion activated parameters, combining Gated DeltaNet linear attention, gated attention, MoE layers, and multi-token prediction. The model provides a native 262K-token context window that can be extended to approximately 1M tokens, supports configurable reasoning effort, and is optimized for complex multi-step task execution, autonomous planning, coding agents, and long-context workloads.
|
grok Grok 4.7 is xAI's frontier reasoning model for coding, long-running agentic tasks, professional knowledge work, engineering, computer use, and complex multi-step workflows. It uses a larger base model than Grok 4.6 and was trained with a longer reinforcement-learning run focused on difficult tasks requiring extended execution and self-verification. Grok 4.7 supports a 500K-token context window, text and image understanding, function calling, structured outputs, configurable reasoning effort, and tool-driven agentic workflows.
|
openai GPT-4.1 is OpenAI's 1.6 trillion parameter multimodal reasoning system featuring a 256-expert MoE architecture with enhanced tool integration and world knowledge grounding. Optimized for enterprise applications with improved factual accuracy and reduced hallucination.
|
elevenlabs Eleven Realtime v1 Max is ElevenLabs' live speech-to-speech model, the larger size, for harder questions and multi-step tool use. Connect to WS /v1/realtime, stream 8, 16 or 24 kHz PCM (or 8 kHz G.711) microphone audio in and get 24 kHz spoken replies back, using the OpenAI Realtime event format. Speaks 90+ languages with ElevenLabs voices and supports function tools. Billed per connected minute plus the LLM tokens each conversation uses.
|
openai A large, multimodal language model with advanced reasoning capabilities and support for tools. It supports the new OpenAI Responses API with features like reasoning effort control.
|
openai OpenAI's GPT-5.1 model, released November 13, 2025. Features a strong multimodal and reasoning capabilities, and is optimized for complex agentic workflows. It introduced significant improvements in coding, reasoning, and tool use over previous generations.
|
anthropic Claude Sonnet 5 is the most agentic Sonnet model yet, released by Anthropic on June 30, 2026. It offers a 1M token context window and 128k max output tokens. It is designed for complex, multi-step agentic tasks like coding, tool use, and autonomous planning, achieving performance close to the more expensive Opus 4.8 model. It features adaptive thinking enabled by default and a new tokenizer that produces approximately 30% more tokens for the same text compared to Sonnet 4.6.
|
openai GPT-6 Sol is OpenAI's GPT-6 model designed to balance advanced intelligence, reasoning capability, and cost for demanding production workloads. It is built for complex coding, agentic workflows, professional knowledge work, multimodal understanding, and multi-step task execution. The model supports a 1.05M-token context window, up to 128K output tokens, configurable reasoning effort, image understanding, function calling, structured outputs, web search, file search, and computer use.
|
anthropic Claude Sonnet 5.5 is Anthropic's fast, highly capable hybrid reasoning model designed for coding, agentic workflows, computer use, knowledge work, and everyday professional tasks. It improves substantially over Sonnet 5 while generating outputs more than 30% faster and using fewer tokens to complete comparable work. The model supports configurable reasoning effort, long-running tool-driven workflows, image understanding, computer interaction, and complex multi-step execution. It is particularly suited to software engineering, iterative development, enterprise automation, document and spreadsheet creation, research, and high-volume agentic applications where a balance of capability, speed, and cost is important.
|
openai GPT-6.1 Sol is OpenAI's advanced reasoning model designed to deliver near-GPT-6 Astra capability at a lower cost for complex coding, computer use, agentic workflows, and professional knowledge work. It supports a 1.05M-token context window and up to 128K output tokens, enabling it to work across large codebases, documents, and long-running multi-step tasks. The model supports configurable reasoning effort, image understanding, function calling, structured outputs, web and file search, code execution, computer use, hosted shell, MCP, and other agentic tools through the Responses API. GPT-6.1 Sol is optimized for applications that require strong reasoning and tool use while balancing capability, speed, and serving cost.
|
novita Kimi K3.1 Pro is Moonshot AI's flagship model, a post-trained version of Kimi K3 with much more reinforcement learning, and its most capable open-weight model to date. It improves on K3 in coding, 3D creation, financial research and agentic work, always reasons (reasoning_effort low, high or max), reads images, calls tools and has a 1M-token context window.
|
openai GPT-4o (Omnimodal) is OpenAI's 1.2 trillion parameter multimodal foundation model featuring unified input processing across text, vision, and audio. Optimized for real-time interaction with enhanced reasoning and cross-modal understanding capabilities.
|
openai GPT-5.6 Terra is a balanced model in OpenAI's GPT-5.6 series, the balanced workhorse tier of the GPT-5.6 family. Delivers highly reliable performance for everyday professional workflows, code generation, and general-purpose agentic tasks at a lower cost than Sol.
|
openai The standard 'Thinking' variant of OpenAI's GPT-5.2 model, released Dec 11, 2025. Excels at complex reasoning, coding, and multi-step projects. Optimized for professional knowledge work and serves as the balanced option between speed and capability in the GPT-5.2 family.
|
openai GPT-5.4 is OpenAI's flagship frontier model built for sustained, multi-step reasoning with reliable follow-through. It excels at agentic workflows, research, document analysis, and powering complex internal tools with robust instruction following.
|
anthropic Claude Sonnet 4.6 is Anthropic's balanced production model, offering near-Opus level intelligence at a much lower cost. It features a full upgrade across coding, computer use, long-context reasoning, agent planning, knowledge work, and design, along with an adaptive thinking capability.
|
novita Kimi’s most capable flagship model to date, with 2.8 trillion parameters. Built for frontier intelligence scenarios including long-horizon coding, knowledge work, and reasoning. The world’s first open-source model in the 3-trillion-parameter class.
|
openai GPT-5.6 Sol is the flagship model in the GPT-5.6 family, designed for the most complex reasoning, agentic coding, and deep research tasks. It features a 1.05M token context window and excels in long-horizon, multi-step workflows. It supports advanced features like programmatic tool calling, multi-agent orchestration, and a 'max' reasoning effort
|
anthropic Claude Opus 5.5 is Anthropic's advanced Opus model for complex coding, long-running autonomous agents, computer use, scientific research, knowledge work, and professional workflows. It delivers substantial improvements in agentic coding, software engineering, tool use, visual reasoning, and long-horizon task execution while using fewer tokens and less compute than Opus 5. The model supports adaptive thinking with configurable reasoning effort, function and tool use, computer-use workflows, vision understanding, and long-context agentic applications.
|
novita Kimi K3.1 Pro Highspeed is the same model as Kimi K3.1 Pro, served by Moonshot AI at two to three times the output speed for more responsive use. It always reasons (reasoning_effort low, high or max), reads images, calls tools and has a 1M-token context window.
|
anthropic Claude Opus 4.6 brings Anthropic's advanced reasoning capabilities to high-stakes workflows. With full adaptive thinking support and a 1-million-token context window, it excels at complex tasks across coding, cybersecurity, financial analysis, and large-scale enterprise agents.
|
anthropic Claude Opus 4.7 excels in autonomous multi-step engineering and long-horizon agentic workflows. It features an Effort Control layer for self-correction and advanced visual intelligence, making it an ideal choice for software engineering, legal, and financial analyses.
|
anthropic Claude Opus 4.8 is a workhorse flagship model designed for high-end reasoning and long-horizon agentic coding. It offers exceptional consistency and autonomy to sustain work on complex, long-running tasks.
|
anthropic Claude Opus 5 is Anthropic's flagship model designed for complex agentic coding and enterprise work, delivering intelligence close to Claude Fable 5 at half the price. It features a 1M token context window, adaptive thinking, and state-of-the-art performance on coding and knowledge work evaluations. It is optimized for bounded, complex tasks
|
anthropic Claude Opus 4.5 is Anthropic's Opus model from November 2025, for complex reasoning, coding and agentic work. It reads text and images, supports extended thinking and the effort setting, calls tools and has a 200K-token context window.
|
openai GPT-5.5 is a highly capable and efficient model featuring dynamic routing between 'Instant' and 'Thinking' modes. It provides improvements in conceptual clarity, scientific research ability, accuracy during knowledge work, and polished conversational tones.
|
openai Chat Latest points to the Instant model OpenAI currently uses in ChatGPT, and OpenAI updates the model behind it regularly. It reads text and images, calls tools and has a 400K-token context window.
|
anthropic Claude Fable 5 is a Mythos-class model from Anthropic, built for autonomous knowledge work and coding. It supports text, image, and file inputs with text output, with reasoning support and a 1M-token context window. It is suited for long-running, complex, and asynchronous tasks that previously required frequent human check-ins.
|
anthropic Claude Fable 5.1 is Anthropic's frontier intelligence model designed for demanding reasoning, long-horizon agentic work, advanced software engineering, multistep research, and complex enterprise workflows. It is built for tasks that can span hours and multiple applications, with strong planning, tool orchestration, failure recovery, coding, document understanding, and vision capabilities. The model supports a 1M-token context window, adaptive reasoning that is always enabled, and up to 128K output tokens.
|
openai GPT-6 Astra is OpenAI's flagship model, released 3 September 2026. It costs $10 per million input tokens and $50 per million output, with a 1,050,000-token context window and 128,000 max output tokens. It accepts text and images, and supports streaming, function calling and structured outputs.
|
openai OpenAI's flagship GPT-5 Pro model, released October 6, 2025. A highly capable multimodal model featuring advanced reasoning, native tool use, and strong performance on complex professional and coding tasks. It introduced a structured 'reasoning effort' parameter for controlling response depth.
|
openai OpenAI's flagship GPT-5.2 Pro model, Designed for professional knowledge work and agentic workflows, exclusive 'xhigh' reasoning effort, and top-tier performance on complex reasoning, coding, and scientific benchmarks.
|
openai GPT-5.4 Pro offers deeper, higher-reliability reasoning for complex production scenarios. It is engineered for high-stakes agentic workflows, long-form analysis and synthesis, complex planning, and advanced reasoning tasks where accuracy is paramount.
|
openai GPT-5.5 Pro is the highest-capability model in the 5.5 lineup, optimized for the hardest tasks and long-running workflows. It uses significant compute to 'think' harder before answering, making it ideal for deep research, highly complex coding, and extensive multi-step problem solving.
|
 Gemini 3 Pro Image, also known as Nano Banana Pro, is Google's professional image generation and editing model with a reasoning core. It is designed for studio-quality visuals, complex compositions and layouts, accurate text rendering, and high-resolution creative production.
|
Gemini 3.1 Flash Image, also known as Nano Banana 2, is Google's high-efficiency native image generation and conversational editing model. It is optimized for low-latency, high-volume image workloads and supports high-quality image generation, editing, improved text rendering, multiple aspect ratios, search-informed generation, and resolutions up to 4K.
|
Gemini Omni 1.1 Flash is Google's high-performance multimodal video generation and editing model designed for fast, conversational video creation and cinematic control. It supports text-to-video, image-to-video, video-to-video editing, natural-language conversational editing, video extension, first-and-last-frame interpolation, and resolution upscaling. The model combines multimodal understanding with generative video capabilities to produce coherent, controllable video with synchronized audio. It supports a 1,048,576-token context window and generates videos lasting 3 to 10 seconds at 24 FPS, with output resolutions ranging from 360p to 4K. It is suitable for filmmaking, advertising, animation, social media content, creative production, and iterative video editing workflows.
|
Gemini Omni Flash is Google's multimodal video generation and editing model designed for fast, conversational video creation. It supports text-to-video, image-to-video, video-to-video editing, video extension, keyframe interpolation, resolution upscaling, and natural-language iterative editing through the Gemini Interactions API. The model generates videos with synchronized audio and supports a 1,048,576-token context window. It can generate videos lasting 3 to 10 seconds at 24 FPS, with supported output resolutions ranging from 360p to 4K. It is suitable for advertising, filmmaking, animation, social media production, and creative video workflows.
|
Lyria 3.5 is Google's advanced music generation model designed for creating high-quality, full-length songs from text prompts and image inputs. It produces structurally coherent music with multiple verses, choruses, bridges, vocals, lyrics, and complete instrumental arrangements, with controllable track duration up to approximately three minutes. The model supports detailed creative direction over genre, instrumentation, tempo, dynamics, vocal style, acoustic characteristics, and song structure, while also allowing users to provide their own lyrics or request instrumental-only music. Lyria 3.5 generates professional-grade 44.1 kHz stereo audio and is designed for songwriting, soundtrack creation, advertising, content production, music prototyping, and other professional creative workflows.
|
2d0055472a04402b964549051f5916f7 Codestral 25.08 is Mistral AI's code model for low-latency, high-frequency coding tasks such as code generation and completion. It calls tools and has a 128K-token context window.
|
c76881da022946739a627360e94fcf09 Ministral 3 14B is the largest model in Mistral AI's Ministral 3 family, built for local and edge deployment with performance close to Mistral Small 3.2 24B. It reads text and images, calls tools and has a 256K-token context window.
|
8c7155fe8f264cf386848f276ebe8f21 Ministral 3 3B is the smallest model in Mistral AI's Ministral 3 family, a compact text and vision model for low-cost, low-latency tasks. It calls tools and has a 256K-token context window.
|
f83ee0cff9ef437e917ee29b0f5be9c5 Ministral 3 8B is the mid-size model in Mistral AI's Ministral 3 family, an efficient text and vision model for edge deployment. It calls tools and has a 256K-token context window.
|
6dda3aa353094e2eb45d552afe4eafdb Mistral Small 4 is Mistral AI's hybrid model that brings instruct, reasoning and coding together in one 119B mixture-of-experts model with 6.5B active parameters. It reads text and images, calls tools, thinks step by step when asked (reasoning_effort high, or none to answer at once) and has a 256K-token context window.
|
grok Grok Voice 3.0 is xAI's live speech-to-speech model, released in October 2026. Connect to WS /v1/realtime, stream microphone audio in and get spoken replies back with low latency, using the OpenAI Realtime event format. Supports function tools and sessions of up to 120 minutes. Billed per connected minute, plus a small fee per text message.
|
How to choose
- Conversational applications
- AI agents
- Code generation
- RAG and document analysis
Pricing methodology
OneInfer compares only positive prices with the same billing unit. Per-minute, per-character, per-token, per-image, per-video, and per-second rates remain separate. Prices can change, so the current model page and console remain the source of truth.
Frequently asked questions
How are text model prices compared?
When input and output token prices are available, OneInfer uses their average as a simple comparison score. The full input and output rates remain visible on each model page.
Can I switch text models without changing my integration?
Yes. Models are accessed through the unified OneInfer API; change the model identifier while keeping the same authentication and request pattern.