Claude Fable 5
Anthropic's most capable generally available model — a Mythos-class model made safe for general use. #1 on the Artificial Analysis Intelligence Index (64.9), 80.3% on SWE-Bench Pro, 1M-token context with always-on adaptive thinking. Released June 9, 2026.
Model Specs
- Released
- Jun 2026
- Context window
- 1.0M tokens
- Max output
- 128K tokens
- Capabilities
- reasoningmultimodalfunction-callinglong-context
- Modalities
- textvision
About this model
Claude Fable 5 is Anthropic's most capable generally available model, released June 9, 2026 — positioned as a new tier above the Opus line. Anthropic describes it as "a Mythos-class model made safe for general use": the same underlying model as the restricted Claude Mythos 5, with safety guardrails built in for broad availability. On the Artificial Analysis Intelligence Index it scores 64.9 — the #1 ranked model — ahead of GPT-5.5 (60) and every Gemini variant.
The coding results define the release. Fable 5 scores **80.3% on SWE-Bench Pro**, far ahead of Claude Opus 4.8 (69.2%), GPT-5.5 (58.6%), and Gemini 3.1 Pro (54.2%). On Cognition's FrontierCode Diamond tier — 150 real-world pull requests at the hardest difficulty — it reaches 29.3%, more than double Opus 4.8's 13.4%. On GDPval-AA (real-world economic tasks) it leads with an Elo of 1932, and it posts 53% on Humanity's Last Exam. Reasoning is handled by always-on **adaptive thinking**: the model decides how much to deliberate per request, tuned through an effort parameter (low → max) rather than manual thinking budgets. Classic sampling controls (temperature, top_p, top_k) are removed entirely.
The context window is 1M tokens (using the newer Opus 4.7 tokenizer — the same text yields roughly 30% more tokens than pre-4.7 models) with up to 128K output tokens. Raw API pricing is $10/M input and $50/M output, with a 90% prompt-caching discount ($1.00/M on cache hits) — less than half the price of the earlier Mythos Preview. Reach for Fable 5 when the task justifies frontier capability: hard multi-file engineering work, deep research and analysis, agentic workflows that run for hours, and high-stakes writing. For routine work, Claude Sonnet 4.5 remains the faster, cheaper default.
Key Strengths
#1 on the Intelligence Index
Artificial Analysis ranks Fable 5 first among all models with an Intelligence Index of 64.9 — ahead of GPT-5.5 (60) and Gemini 3.5 Flash (55). The composite spans agents, coding, general capability, and scientific reasoning.
State-of-the-art coding (SWE-Bench Pro 80.3%)
80.3% on SWE-Bench Pro versus 69.2% for Opus 4.8 and 58.6% for GPT-5.5, plus 29.3% on Cognition's FrontierCode Diamond — more than double the next-best model on real-world hard PRs.
Always-on adaptive thinking
Reasoning is built in, not opt-in. The model adapts deliberation depth per request, controlled by an effort parameter (low / medium / high / xhigh / max) instead of manual thinking-token budgets.
1M-token context
Full-codebase and multi-document workflows in a single request. Uses the Opus 4.7 tokenizer (same text yields ~30% more tokens than older Claude models), with up to 128K output tokens.
Real-world economic task leader
GDPval-AA Elo of 1932 — the highest of any model (GPT-5.5: 1769) — measuring performance on tasks drawn from real occupations, plus the top reported score on the Hebbia Finance Benchmark.
90% prompt-caching discount
Cache hits cost $1.00/M tokens against a $10/M base input rate. For agentic and long-context workflows with repeated prefixes, effective input cost drops dramatically.
How it compares
Claude Fable 5 sits at the top of the frontier tier. Compare against the alternatives based on task difficulty, budget, and latency tolerance.
| vs. Model | Verdict | Outcome |
|---|---|---|
| Claude Opus 4.1 | Fable 5 is a full tier above: SWE-Bench Pro 80.3% (Pro) vs Opus's SWE-bench Verified 74.5%, 1M vs 200K context, always-on adaptive thinking vs manual extended thinking — and at $10/$50 raw it's actually cheaper than Opus 4.1's $18.75/$75. Opus 4.1 remains a strong premium workhorse, but Fable 5 wins for frontier work. | Wins most cases |
| GPT-5.5 | Fable 5 leads on intelligence (AA 64.9 vs 60), coding (SWE-Bench Pro 80.3% vs 58.6%), and real-world tasks (GDPval-AA 1932 vs 1769 Elo). GPT-5.5 is half the price ($5/$30 vs $10/$50) with comparable context and a strong agentic/computer-use story. Pick Fable 5 for the hardest problems; GPT-5.5 for frontier-adjacent work at lower cost. | Wins most cases |
| Claude Sonnet 4.5 | Sonnet 4.5 is the everyday default: far cheaper ($3.75/$15 raw), much faster to first token (1.7s vs ~108s), and excellent at routine coding and writing. Fable 5 justifies its premium only when the task is genuinely hard — frontier coding, deep research, long-horizon agents. Most workflows should mix both. | Depends |
Pros
- #1 model on the Artificial Analysis Intelligence Index (64.9)
- State-of-the-art coding: SWE-Bench Pro 80.3%, FrontierCode Diamond 29.3%
- 1M-token context with 128K output
- Always-on adaptive thinking — no manual reasoning budgets
- Cheaper than Claude Opus 4.1 raw rates despite being far more capable
- 90% prompt-caching discount ($1.00/M cache hits)
- Top reported scores on GDPval-AA and Hebbia Finance
Things to consider
- Premium pricing — $10/M input, $50/M output raw
- Slow time-to-first-token (~108s on hard tasks) due to always-on reasoning
- Knowledge cutoff not publicly disclosed by Anthropic
- Sampling controls (temperature/top_p/top_k) removed — less generation-level tuning
- Sensitive cyber/bio topics fall back to Claude Opus 4.8 in <5% of sessions
- Overkill for routine tasks where Sonnet 4.5 is faster and cheaper
Best use cases
Hard software engineering
Multi-file refactors, legacy migrations, and real-PR-grade fixes. The SWE-Bench Pro and FrontierCode results translate directly to fewer iterations on genuinely difficult engineering tasks.
Long-running agentic workflows
Always-on adaptive thinking plus 1M context make Fable 5 suited to agents that plan, call tools, and operate over hours — research agents, code agents, and pipeline orchestrators.
Deep research & analysis
53% on Humanity's Last Exam and the top GDPval-AA score signal strong performance on expert-level reasoning — literature reviews, technical due diligence, scientific analysis.
Full-codebase comprehension
Load an entire repository into the 1M-token window and ask architecture-level questions, trace bugs across modules, or generate documentation with full context.
High-stakes professional writing
Contracts, technical specifications, and reports where precision matters more than speed. Adaptive thinking spends extra deliberation exactly where the text is hardest.
Financial analysis
Anthropic reports Fable 5 achieved the highest score of any model on the Hebbia Finance Benchmark — document-heavy financial reasoning across filings, models, and memos.
How to use it on Renas AI
- 1
Step 1
Open AI Chat on Renas
Navigate to AI Chat in the Renas dashboard. Claude models appear in the model selector — pick the most capable Claude available to your plan.
- 2
Step 2
Give it the full problem
Fable-class models reward complete context. Paste entire files, long documents, or multi-part task descriptions — the 1M-token window handles what older models had to truncate.
- 3
Step 3
Let adaptive thinking work
There's no thinking toggle to manage — the model deliberates automatically and goes deeper on harder requests. Expect slower first tokens on complex tasks; the quality difference is the point.
- 4
Step 4
Iterate within one conversation
Long context means revisions, follow-ups, and side-questions stay grounded in everything discussed so far. One thread can carry an entire project phase.
Pricing
Pricing on Renas AI
Pay-as-you-go credits, no API keys, no rate limits.
~25,000 words on a 10,000-credit Spark plan
Frequently asked questions
Other Anthropic models
Other text models on Renas AI
Frontier intelligence on Renas
Access leading Claude models with your Renas AI subscription credits — no API key, no setup, no per-seat fees.
Try Claude on Renas