Claude Fable 5.1 vs Opus 5.5: which should you use?

Claude Opus 5.5 is the better default for most coding work, and it costs 60% less per token than Claude Fable 5.1: $4 per million input tokens and $20 per million output, against $10 and $50. In the Opus 5.5 launch post, Anthropic says Opus 5.5 "performs at the level of Claude Fable 5.1 on most work" and leads it on agentic coding, computer use and knowledge-work benchmarks, including Terminal-Bench 4.0 (66.4% to 55.8%). Independent results agree: Vals AI's Terminal-Bench 4.0 run puts Opus 5.5 at 65.2% and Fable 5.1 at 58.1%, and people's votes on LMArena's WebDev leaderboard rank Opus 5.5 first and Fable 5.1 fifth. Anthropic's own docs say to start with Opus 5.5 and move to Fable 5.1 for demanding reasoning and long-horizon agent work, or when Opus at higher effort still falls short.

Key takeaways

• Price: Fable 5.1 is $10 / $50 per million tokens and Opus 5.5 $4 / $20. Both have a 1M-token context window with no long-context surcharge, and 128,000 output tokens.
• Benchmarks: Opus 5.5 is ahead on every benchmark in its launch table. The gap is widest on Terminal-Bench 4.0, at 10.6 points, and narrowest on computer use.
• Which first: Anthropic recommends Opus 5.5 for most workloads and Fable 5.1 when Opus at higher effort isn't enough.
• Plans: Opus 5.5 is included from Claude Pro up. Fable 5.1 costs usage credits on Pro and standard Team seats, and can use up to half the weekly limits on Max and premium seats.

Sources: Anthropic's Opus 5.5 launch post, What's new in Claude Fable 5.1, API pricing and Fable on your plan; Vals AI and LMArena, October 26–27, 2026.

What are Fable 5.1 and Opus 5.5?

Claude Fable 5.1, released September 1, 2026, is Anthropic's most capable generally available model and the successor to Fable 5. Anthropic describes it as built for long-running agentic coding, knowledge work and research. It was released alongside Claude Mythos 5.1, the same model with fewer safeguards, which only organizations verified through Anthropic's cyber and life-sciences programs can use.

Claude Opus 5.5, released three weeks later on September 22, is the next version of Anthropic's Opus line. It's the default model in Claude Code on paid plans and the API, and Anthropic's model guide makes it the place to start when you're unsure which model to use.

The order of release explains the odd result: the newer Opus caught up with an older, larger model at less than half the price.

How much do they cost?

Per million tokens on Anthropic's API:

• Fable 5.1: $10 input, $50 output. Cache reads are $0.25 and cache writes $12.50 (5-minute) or $20 (1-hour). Batch processing halves input and output to $5 and $25.
• Opus 5.5: $4 input, $20 output. Cache reads are $0.20 and cache writes $5 or $8. Batch is $2 and $10. A fast mode, up to 2.5 times the speed, costs $8 and $40; Fable has no fast mode.

Both bill the full 1M-token window at the standard rate, and US-only inference adds 10% to either. Fable's cache reads cost 0.025 times its input price, the lowest ratio of any Claude model, which narrows the gap in long agent sessions that keep re-reading the same context. Details in Claude Fable pricing.

Per task, the gap is smaller than the 2.5x per-token difference. On Vals AI's Terminal-Bench 4.0 run (updated October 26, 2026), an average Fable 5.1 task cost $17.18 and an Opus 5.5 task $13.20.

How do they compare on benchmarks?

From the table in Anthropic's Opus 5.5 post, both at max effort unless noted:

• Terminal-Bench 4.0 (agentic coding): Opus 5.5 66.4% (at xhigh), Fable 5.1 55.8%.
• FrontierCode 1.1, Main: 54.4% against 50.3%.
• CursorBench 4.0: 57.8% against 51.8%.
• AutomationBench (business workflows): 40.0% against 31.4%.
• GDPval-AA v2.1 (knowledge work): 1846 against 1735 Elo.
• Humanity's Last Exam, with tools: 67.7% against 65.6%.
• OSWorld 2.1, partial (computer use): 81.8% against 80.7%.

Anthropic adds that "the gap between Opus 5.5 and Claude Fable 5.1 is narrower than these scores suggest," and that Opus 5.5's main advantage is efficiency: fewer tokens per task, lower prices and faster output.

Independent results:

• Vals AI, Terminal-Bench 4.0 (one shared harness, October 26): Opus 5.5 65.15% in about 1 hour 4 minutes per task, Fable 5.1 58.08% in about 1 hour.
• LMArena WebDev (people voting on real web builds, October 27): Opus 5.5 at max effort is first with 1809, and Fable 5.1 at max effort fifth with 1745.

When is Fable 5.1 worth it?

Anthropic's Fable 5.1 docs say: "For most workloads, start with Claude Opus 5.5." They point to Fable 5.1 for:

• Demanding reasoning and long-horizon agent work, such as multi-file features, large refactors and migrations, and debugging across sessions that run for hours.
• Work where Opus 5.5 at higher effort still falls short on your own evaluations.
• Documents, spreadsheets and slides built from a blank page, and multistep web research, two areas where Anthropic says Fable 5.1 improved most over Fable 5.

The benchmarks above are averages over many tasks. If a specific hard problem has stumped Opus 5.5 at high effort, trying Fable 5.1 on it is the documented next step.

What else is different?

Fable 5.1 has API differences that Opus 5.5 doesn't:

• Thinking is always on, and effort sets how much. Claude Code runs Fable 5.1 at high effort by default and claude.ai at medium. Opus 5.5 defaults to medium in Claude Code and the API.
• Forced tool use returns an error. Developers who set tool_choice to a specific tool need strict tool use or structured outputs instead.
• Its reasoning stays with it. Only Fable 5.1 and Mythos 5.1 can read Fable 5.1's thinking blocks, so a conversation that moves from Fable 5.1 to another model loses that reasoning for the turns that run there.
• Data retention: Fable 5.1 carries 30-day retention and isn't available under zero data retention unless Anthropic expressly authorizes it.
• Safeguards: Fable 5.1 can help find software vulnerabilities but not write exploits. Its classifiers can return a refusal, and some refusals before any output are billed (since September 24, 2026).

Which Claude plans include each?

• Free: neither. Claude's free plan has Sonnet and Haiku.
• Pro ($20 a month, $17 billed annually): Opus 5.5 is included. Fable 5.1 is billed to pay-as-you-go usage credits from the first message.
• Max ($100 or $200 a month) and premium Team and Enterprise seats: both. Fable can use up to 50% of the weekly limit at no extra cost, then usage credits.
• Standard Team and Enterprise seats: Opus 5.5 is included, and Fable 5.1 uses credits.

In Claude Code, Fable 5.1 needs version 2.1.255 or later; switch with /model. More on plan limits in Claude Max vs Pro.

Can you use both in one place?

In Claude's apps and Claude Code you switch with the model picker. In Poly (usepoly.co), Opus 5.5 and Fable 5.1 are both on paid plans, and each person in a shared room picks their own model for their own turns. One person can plan with Fable 5.1 while another builds with Opus 5.5, or Sonnet 5.5 on the free tier, in the same conversation. For a cheaper pair, see Sonnet 5.5 vs Opus 5.5; for OpenAI's flagship, Fable 5.1 vs GPT-6 Astra.