Benchmark scoreboard
Nine evaluations, side by side. Fable 5.1 leads on every row.
| Benchmark | Fable 5.1 | Fable 5 | Delta |
|---|---|---|---|
| Terminal-Bench-Science 0.1 | 52.6% | 24.7% | +27.9 |
| Terminal-Bench 4.0 | 55.8% | 42.0% | +13.8 |
| CursorBench 3.2.0 | 73.4% | 70.5% | +2.9 |
| OSWorld 2.0 (partial) | 77.9% | 72.9% | +5.0 |
| OSWorld 2.0 (strict) | 41.7% | 36.1% | +5.6 |
| HLE (no tools) | 60.9% | 57.8% | +3.1 |
| HLE (with tools) | 65.0% | 63.8% | +1.2 |
| AutomationBench | 31.4% | 17.1% | +14.3 |
| GDPval-AA v2 | 1853 | 1723 | +130 |
The cache arithmetic on three profiles
Cost equals uncached input times the input rate plus cached input times the cache-read rate plus output times the output rate. Cache WRITE cost is excluded because a long-running agent writes the prefix once and reads it many times.
| Profile | Input tokens | Hit rate | Output tokens | Fable 5.1 | Fable 5 | Saving |
|---|---|---|---|---|---|---|
| A. Chat / short prompt | 200K | 20% | 20K | $2.61 | $2.64 | 1.1% |
| B. Typical agentic session | 10M | 90% | 200K | $22.25 | $29.00 | 23.3% |
| C. Long-running cache-heavy agent | 20M | 95% | 100K | $19.75 | $34.00 | 41.9% |
Should you upgrade?
On profile A the saving is about 1% and not worth a migration. On profiles B and C the upgrade pays back on the first session. Anthropic's claim of "about 25% cheaper for typical workloads, up to about 45% for agentic work" reproduces from the published rates alone.
Ready to test the workflow?
Create account & add creditsMigration
Migrating from claude-fable-5 to claude-fable-5.1 is a one-line model-string change. Two behaviour caveats: thinking blocks are model-specific and cannot be replayed into a different model; accounts created on or after 31 August 2026 cannot edit prior context mid-conversation.
Is Claude Fable 5 going away?
No. Anthropic has not published an end-of-life date for Claude Fable 5. The 1 September 2026 release of Claude Fable 5.1 is additive rather than a replacement, and both model identifiers, claude-fable-5 and claude-fable-5.1, resolve today. Pricing on Claude Fable 5 is unchanged at $10 per 1M input and $50 per 1M output tokens, with cache reads at $1.00 per 1M.
Frequently asked questions
Which is cheaper, Claude Fable 5.1 or Claude Fable 5?
It depends entirely on how much cached context you re-read per token generated, because list prices are identical at $10 per 1M input and $50 per 1M output. The only difference is the cache read rate, $0.25 against $1.00. On a short prompt at a 20% hit rate the saving is about 1%; on a long-running agent at a 95% hit rate the saving is about 42%.
Should you upgrade from Claude Fable 5 to 5.1?
Whether you should upgrade from Claude Fable 5 to 5.1 turns on how much cached context you re-read per token generated. On a chat-shaped workload the saving is about 1%; on a long-running agent it reaches about 42%. Anthropic's claim of "about 25% typical, up to 45% agentic" reproduces from the published rates alone.
Put Claude Fable 5.1 to work
Fund a controlled evaluation, start with a prepared prompt, and measure quality and cost on your own workload.