Claude Fable 5
Anthropic's most capable widely released model - frontier intelligence for long-running agents.
Context window
1M
Input / 1M tokens
$10.00
Output / 1M tokens
$50.00
Provider
Anthropic
Priced at $10/$50 per 1M input/output tokens - less than half the price of the earlier Claude Mythos Preview. Uses the Opus 4.7-era tokenizer (~30% more tokens than pre-4.7 models for the same text). 1M context; 128k max output. · Data verified 2026-07-02
Claude Fable 5 (June 2026) is a Mythos-class model made safe for general use and Anthropic's most capable generally available model, showing state-of-the-art results on nearly all tested benchmarks - reportedly ~95% on SWE-bench Verified. Its lead over other Claude models grows the longer and more complex the task. It ships with real-time classifiers that redirect a small share of sensitive queries (cyber, bio/chem, distillation) to Opus 4.8; Anthropic states more than 95% of sessions involve no fallback. Adaptive thinking is always on.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Generally available on the Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud, and Microsoft Foundry as of June 9, 2026. Availability was briefly affected by U.S. export controls that were lifted around July 1, 2026; it is again available with strengthened guardrails. On consumption-based Enterprise plans it is fully available; subscription-plan (Pro/Max/Team) access is being restored as capacity permits.
Strengths
- ✓Highest available Claude capability: state-of-the-art on nearly all tested benchmarks (~95% SWE-bench Verified)
- ✓Exceptional long-horizon agentic performance - lead widens on longer, harder tasks
- ✓Frontier software engineering, scientific research, knowledge work, and vision
- ✓1M context with 128k output and always-on adaptive thinking
- ✓Top scores on external evals (Hebbia Finance, Cognition FrontierCode, ~92.6% GPQA Diamond)
Best for developers who...
When to choose it (and when not to)
Reach for Claude Fable 5 when...
- →You need the absolute highest capability Anthropic offers for hard tasks
- →You run long, complex autonomous agents where quality compounds over many steps
- →You do frontier-level engineering, research, or analytics and cost is secondary
- →You want maximum accuracy on the toughest coding and reasoning benchmarks
Look elsewhere if...
- ✕You are cost-sensitive - at $10/$50 it is the priciest current Claude; Sonnet 5 or Opus 4.8 are far cheaper
- ✕Your workloads touch topics that trigger the safety fallback (some sessions get an Opus 4.8 response instead)
- ✕You need guaranteed uninterrupted subscription-plan access, which is still being restored
How to use it
- ›Reserve it for genuinely hard, long-horizon tasks where its lead over Sonnet 5/Opus 4.8 justifies the price
- ›Adaptive thinking is always on - use the effort parameter to tune depth rather than toggling thinking
- ›Handle stop_reason:'refusal' and fallback responses gracefully in production
- ›Recount token budgets under the Opus 4.7-era tokenizer (~30% more tokens)
Quickstart
Pythonimport anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-fable-5",
max_tokens=16000,
output_config={"effort": "high"},
messages=[
{"role": "user", "content": "Implement this multi-service feature end to end."}
],
)
print(response.content[0].text)Adaptive thinking is always on. A small share of sensitive queries may be answered by Claude Opus 4.8 instead; check stop_details/refusals in production.
API model id: claude-fable-5
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| SWE-bench Verified | 95.0% | Per secondary benchmark aggregators (MorphLLM); Anthropic reports state-of-the-art but publishes charts, not this exact figure |
| Terminal-Bench 2.1 | 88.0% | |
| GPQA Diamond | 92.6% | |
| SWE-bench Pro | 80.3% |
Source: MorphLLM Claude benchmark table + Anthropic announcement