Skip to content
FrankX.AI
AnthropicGA

Claude Fable 5

Mythos-class made generally available — the new agentic-coding ceiling, at 2× Opus pricing.

Read the full Claude Fable 5 analysis

Context

1M

Max output

128K

Input /1M

$10.00

Output /1M

$50.00

Best for

  • Agentic pipelines feeding schemas, tools, and other agents (measured constraint precision)
  • Long-horizon coding — SWE-Bench Verified 95% / Pro ~80% at launch (vendor-claimed)
  • Hard reasoning under strict output contracts

Watch out

$10/$50 is double Opus 4.8 standard. Launch benchmarks are vendor-claimed. In our stress round it executed a governance-gated edit without flagging it — pair with structural gates, and enforce output contracts in schemas: every model’s discipline degrades under heavy task load.

For creators. The default Claude Code driver for agentic builds. Route judgment-heavy review and human-read prose to Opus 4.8 at half the price — our blind style verdicts flipped between rounds, so prose is not the upgrade case.

Benchmarks

swe bench verified95
swe bench pro80
cursorbench max effort72.9

Capabilities

  • Mythos-class capabilities made generally available (safety classifiers attached)
  • Default model in Claude Code (claude-fable-5)
  • Leads FrontierCode Diamond and Main subsets at launch
  • Lead widens as tasks get longer and more complex (per Anthropic)
  • 1M context, 128K max output
  • Measured edge (FrankX arena, 4 rounds): output discipline / constraint precision at correctness parity-or-better vs Opus 4.8

Compare Claude Fable 5

Claude Fable 5 vs Claude Opus 4.8

Fable 5 takes agentic coding, constraint precision, and hard reasoning — at double the price. Opus 4.8 keeps situational judgment, code-craft quality, and the better $/token for human-read prose. Route by task shape, not by leaderboard.

Claude Fable 5 vs GPT-5.5

Fable 5 leads agentic coding by a generation-sized margin on launch numbers; GPT-5.5 keeps computer-use, terminal autonomy, and native voice. Different ceilings for different jobs.

Claude Fable 5 vs Gemini 3.5 Pro

Not yet a fair fight: Fable 5 is generally available with published numbers; Gemini 3.5 Pro remains a limited Vertex preview with no model card, benchmarks, or pricing. Today, Fable 5 wins by forfeit — revisit at Gemini GA.

Claude Fable 5 vs Grok 4.3

Different products. Fable 5 is the agentic-coding ceiling; Grok 4.3 is the cheapest credible frontier intelligence with the fastest throughput in its class. The 20× output-price gap means most stacks should run both — at different tiers.

Claude Fable 5 vs DeepSeek V4

Fable 5 owns the ceiling; DeepSeek V4 owns the open-weight floor — 80.6% SWE-Bench Verified under MIT at a tenth of the cost. If sovereignty or self-hosting is a requirement, DeepSeek wins by default; if peak agentic capability is, Fable 5 does.

Claude Fable 5 vs Kimi K2.6

Kimi K2.6 is the best open-weights model on the neutral index and matches GPT-5.5 on SWE-Bench Pro — at $0.60/$2.50. Fable 5 still clears it by a full tier on agentic coding. Ceiling vs best-value-open: route accordingly.

Claude Fable 5 vs Qwen3.7-Max

Qwen3.7-Max leads its peer group (Kimi, DeepSeek) on hard agentic coding and matches Fable 5 on context — at a quarter of the price. Fable 5 keeps a ~19-point SWE-Bench Pro lead. Value leader vs ceiling, both closed.

More from Anthropic

Sources