Claude Fable 5
Language Available Full comparison ↗ anthropic/claude-fable-5 · by Anthropic
· decoder-only-transformer
Pricing — 1 offering(s)
Input tokens
- $10.00 / 1M tokens (input) Current 2026-06-09 → present
Output tokens
- $50.00 / 1M tokens (output) Current 2026-06-09 → present
Cache hits and refreshes
- $1.00 / 1M tokens (cached input) Current 2026-06-09 → present
5-minute cache writes
- $12.50 / 1M tokens (cache write) Current 2026-06-09 → present
Batch input tokens
- $5.00 / 1M tokens (input) Current 2026-06-09 → present
Batch output tokens
- $25.00 / 1M tokens (output) Current 2026-06-09 → present
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Claude Fable 5 fits into a cost-aware routing setup
See how →Capability profile
How Claude Fable 5 rates across core capability dimensions, with the task-level evidence behind each rating.
SWE-bench Verified (500 human-verified GitHub issues; patch must pass the repo's own tests; no file-location hints): resolves 95% — fails roughly 1 issue in 20. Independent (Vals AI leaderboard; Anthropic's own launch materials cite custom partner evals, not SWE-bench).
1M-token context window at standard pricing.
Top-tier model — highest latency and cost in the publicly available Claude lineup.
$10/$50 per MTok — most expensive publicly available Claude model.
Coding benchmarks
View the full coding benchmark leaderboard →| Benchmark | Score | Harness | Source |
|---|---|---|---|
| swe-bench-verified ↗ | 95.0% 2026-06 | — | 📊 leaderboard ↗ |
| terminal-bench-2-1 | 84.6% 2026-08 | terminus-2 | 🔬 independent ↗ |
| terminal-bench-4-0 | 41.4% 2026-09 | mini-swe-agent | 📊 leaderboard ↗ |
Operator guidance
Route here when Opus 4.8 is not sufficient — the new ceiling for publicly available Claude capability. Adaptive thinking is always on, so there is no effort parameter to tune. Drop to Opus 4.8 ($5/$25) for most agentic workloads where cost matters; Sonnet 5 ($2–3/$10–15) for standard production use.
Use cases
- The most demanding reasoning and long-horizon agentic coding tasks
- Complex multi-step agent workflows where quality dominates cost and latency
- Top-end benchmark and evaluation work
Limitations
- Highest cost and latency of any publicly available Claude model ($10/$50 per MTok)
- Updated tokenizer produces ~30% more tokens than Sonnet 4.6 and earlier — factor into cost comparisons
- Adaptive thinking always on — cannot be disabled; may increase output length and cost vs. non-thinking models