Claude Fable 5.2 vs Claude Opus 5.5

Specs and pricing side by side

API list prices read from the model registry (src/lib/models/registry.ts), verified against Anthropic's pricing documentation on November 11, 2026. Fable 5.1 is the reference column: Fable 5.2 replaces it at 20% lower input and output prices.

Prices and specs re-read from Anthropic's pricing page, the Fable 5.2 model card and the API data-retention docs on 2026-11-11 (Fable 5.2) and 2026-10-08 (Opus 5.5). Fast mode is 2x base on both models, is not available with the Batch API, and prompt-caching and data-residency multipliers apply on top of it.

Benchmarks: Anthropic's own launch table

Twelve rows from Anthropic's November 9, 2026 launch post, Fable 5.2 against Opus 5.5, each model at its highest available effort with safeguards on. These are vendor figures — no independent replication of Fable 5.2 had been published when this page was written.

Source: Anthropic, “Introducing Claude Fable 5.2 and Claude Mythos 5.2”, November 9, 2026. Fable 5.2's FrontierCode 2.0 row is at xhigh effort and its Remote Labor Index row at high effort, as labeled in that table. Three rows from the same table are deliberately excluded here — CyScenarioBench, CVE Rediscovery and de novo binder design — because they score Mythos 5.2 against Opus 5.5 “without safeguards”, so they are not Fable 5.2 numbers; Fable 5.2's cyber and biology safeguards block most of those tasks. No third party had published an independent Fable 5.2 evaluation as of November 11, 2026.

When to use which

Anthropic's own routing advice, plus the cost and compliance consequences of following it.

• Your evals say Opus 5.5 can't do it

  Anthropic's stated rule: start on Opus 5.5 and move up to Fable 5.2 only when your evals on an Opus model at higher effort still miss. The launch post frames Fable as the model for “the most ambitious, open-ended technical work” — multi-day migrations, hard research, hardware and science work.

• You orchestrate long-running agent teams

  The demo that matters most: a team of 1,000 Fable 5.2 agents fully formalised Fermat's Last Theorem in 30 hours; the same 1,000-agent setup on Opus 5.5 completed 26% in 18 hours and barely progressed afterwards. If your workload is many subagents over many hours, that is the gap you are paying for.

• The work is routine agentic coding

  Anthropic calls Opus 5.5 its strongest daily driver and defaults it to medium effort. It also reports Fable 5.2 at high effort scoring about level with Opus 5.5 at max on FrontierCode 2.0 for roughly half the cost per task — a vendor claim, but the direction is clear: if your tasks already pass on Opus 5.5, the 2x list price buys nothing.

• Volume dominates your bill

  At $8/$40 against $4/$20, Fable 5.2 doubles the cost of identical token counts — and it defaults to high effort where Opus 5.5 defaults to medium. For high-volume work, Sonnet 5.5 at $2/$10 covers most of it, with Opus 5.5 for the hard tail.

• You need zero data retention

  Fable 5.2 is a Covered Model: 30-day retention is mandatory and zero data retention is not available unless Anthropic expressly authorises it. Enterprise Frontier Safeguards, in private beta, runs the monitoring inside the customer's own infrastructure and is the documented ZDR-compatible path. Opus 5.5 is not on the Covered Models list, so the model itself does not block a ZDR arrangement.

• You want Fable 5.2 as advisor or orchestrator

  Anthropic positions Fable 5.2 for both roles: smaller models consult it through the advisor tool, or it delegates simpler work to subagents. One constraint — an advisor paired with a Fable 5.2 executor must itself be Fable 5.2 or Mythos 5.2; every other advisor, Fable 5.1 included, returns a 400.

Routing guidance is Anthropic's, published November 9, 2026; the cost and retention consequences are read from the same vendor pages. Your own evals outrank any default, and Fable 5.2 is two days old — treat per-task claims as provisional.

What the 2x premium costs in practice

List-price arithmetic on representative workloads. This is not a measured cost per task — Fable 5.2 has no independent cost-per-task measurement yet.

• Agentic coding, 100 tasks

  A typical agentic-coding task spends roughly 10K input tokens with half of them served from cache and 8K output tokens. Cache reads cost the same on both models, so the premium sits almost entirely on fresh input and output — the worked example below has the arithmetic.

• One long-context research job

  Both models serve the full 1M window at standard pricing — neither carries a long-context surcharge, so a 500K-token prompt scales the bill without changing the 2x ratio.

• Batch and caching stack on top

  The Batch API halves both models and prompt caching stacks on the discounted rate. Because cache hits cost the same on both, a cache-heavy batch pipeline narrows the gap more than the headline 2x suggests.

Worked example — 100 tasks, 10K input tokens each with a 50% cache hit, 8K output tokens each, at list rates: Fable 5.2 ≈ $0.36 per task (≈ $36.10 for 100), Opus 5.5 ≈ $0.18 per task (≈ $18.10 for 100). Ratio ≈ 2.0x. Cache reads cost the same $0.20/MTok on both models, so the premium concentrates on fresh input and output.

Same arithmetic for one long-context job — 500K input tokens and 20K output tokens, no cache: Fable 5.2 ≈ $4.80, Opus 5.5 ≈ $2.40. Both models serve the full 1M window at standard pricing, so a bigger prompt scales the bill but does not change the ratio.

Priced at API list rates from the registry, global routing, no data-residency multiplier — Anthropic's 1.1x inference_geo “us” multiplier applies equally to both models. The cost calculator has the exact math for your volumes, caching and batching.

Open the cost calculator

Migrating from Fable 5.1 (and what breaks)

Fable 5.2 changes the model ID and four behaviours. Moving to Opus 5.5 needs none of this — it shares Fable 5.1's request shape.

• The prefix check now applies to every account

  If your code edits earlier turns — a changed system prompt, reordered messages, a different tool list — sending a Fable 5.2 thinking block back returns 400 “The block is bound to a different conversation”, whatever the account's age. Fable 5.1 only did this on accounts created on or after August 31, 2026. The stopgap is the thinking-binding-controls-2026-08-01 beta header with prefix_mismatch_behavior set to drop_block, at the cost of higher output tokens.

• Only Fable 5.2 and Mythos 5.2 read Fable 5.2 thinking blocks

  Route a conversation from Fable 5.2 to Opus 5.5 and the reasoning is dropped — the request still succeeds and dropped blocks are not billed, but the model loses that context. Moving the other way, from Opus 5.5 to Fable 5.2, keeps it.

• computer_20251124 now returns 400

  Declare computer_toolset_20260801 instead. Amazon Bedrock is the exception: the earlier tool still works there, exactly as it does on Fable 5.1.

• An advisor must match the executor

  A Fable 5.2 executor accepts only Fable 5.2 or Mythos 5.2 as its advisor; anything else, Fable 5.1 included, returns 400. Keeping your current executor and pointing the advisor at claude-fable-5-2 works for every executor model.

• Re-run your effort sweep

  The API default is high. Anthropic's guidance is to start there and step up to xhigh only where you have measured a gain, reserving max for ceiling measurements. Prompt lines that ask for less deliberation no longer suppress thinking at medium effort and above.

Source: Anthropic's “What's new in Claude Fable 5.2” and the Fable 5.2 model card, read November 11, 2026. If Claude Code, claude.ai, Claude Managed Agents or the Claude Agent SDK builds your requests, none of these breaking changes apply.

How this differs from Fable 5 vs Opus 5

This site keeps two comparison pages because they answer different questions.

The older page covers Fable 5 against Opus 5 — the July 2026 generation, where Opus 5 at $5/$25 undercut Fable 5's $10/$50 and Anthropic's advice flipped from Fable-first to Opus-first. It remains accurate for those two models. This page covers the November 2026 generation: Fable 5.2 at $8/$40 against Opus 5.5 at $4/$20, where both models have moved and the gap is still exactly 2x. Choosing between the current models? You want this page. Already on Opus 5 or Fable 5 and deciding whether to move? Read both.

Claude Fable 5 vs Opus 5

Keep reading