Claude Fable 5.2 vs Opus 5.5: start on Opus, move up when your evals say so
Claude Fable 5.2 vs Opus 5.5 is mostly a question of price per task. Both have a 1M token context window and 128K output, both think adaptively, and both learned up to June 2026. Fable 5.2 costs twice as much per token and scores higher on most of Anthropic's benchmarks. Anthropic itself tells most teams to start on Opus 5.5.
Claude Fable 5.2 vs Opus 5.5
- Checked
- 9 November 2026
- Price per MTok
- Fable $8 / $40, Opus $4 / $20
- Default effort
- Fable high, Opus medium
- Context and output
- 1M and 128K on both
- Anthropic's advice
- Start on Opus 5.5
The short answer
Run Opus 5.5 for most coding and knowledge work, and move a task to Fable 5.2 when your own evals at higher effort on Opus still fall short. Fable 5.2 lists at $8 and $40 per million tokens against $4 and $20, defaults to high effort instead of medium, and needs 30-day data retention. Anthropic calls Opus 5.5 its strongest daily driver.
Which model fits which job
Anthropic's announcement says Fable and Mythos are "best for the most ambitious, open-ended technical work" and recommends Opus 5.5 as the starting point for most coding and knowledge tasks. Its what's new page says to move to Fable 5.2 when evals on an Opus model at higher effort still miss your requirements. The table applies that to common jobs.
| Job | Start on | Why |
|---|---|---|
| Daily coding in Claude Code | Opus 5.5 | Anthropic's recommended daily driver, at half the token price. |
| Agent sessions that run for hours | Fable 5.2 | The model page names multihour agent sessions as its typical use. |
| Hardest debugging, migrations, research | Fable 5.2 | It leads Opus 5.5 on most of Anthropic's reported benchmarks. |
| High-volume API traffic | Opus 5.5 | Lower price and a higher rate limit at each tier. |
| Organisations on zero data retention | Opus 5.5 | Fable 5.2 needs 30-day retention unless Anthropic authorises otherwise. |
| Knowledge work scored by people | Either | On AA-Briefcase Elo, Opus 5.5 scores 1800 and Fable 5.2 scores 1785. |
The two models side by side
Every row comes from Anthropic's Fable 5.2 model page, its Opus 5.5 model page and its pricing page as they stood on 9 November 2026.
| Item | Fable 5.2 | Opus 5.5 |
|---|---|---|
| Model ID | claude-fable-5-2 | claude-opus-5-5 |
| Released | 9 November 2026 | 22 September 2026 |
| Context window and max output | 1M tokens and 128K | 1M tokens and 128K |
| Input and output per million tokens | $8 and $40 | $4 and $20 |
| Cache reads per million tokens | $0.20 | $0.20 |
| Batch API per million tokens | $4 and $20 | $2 and $10 |
| Fast mode per million tokens | $16 and $80 | $8 and $40 |
| Default effort on the API | high | medium |
| Thinking | Adaptive, always on | Adaptive, always on |
| Knowledge cutoff | June 2026 | June 2026 |
| Zero data retention | Not available unless authorised | Can run under it |
| Retirement not before | 9 November 2027 | 22 September 2027 |
What Anthropic's benchmarks show
These scores come from the table in Anthropic's announcement. They are the vendor's own numbers on the vendor's chosen tests, so treat them as a reason to run your own comparison, not as the comparison.
| Benchmark | Fable 5.2 | Opus 5.5 |
|---|---|---|
| FrontierCode 2.0, agentic coding | 49.1% at xhigh effort | 45.5% |
| FrontierSWE v2, agentic coding | 69.9% | 62.3% |
| AutomationBench, business workflows | 51.4% | 42.5% |
| OSWorld 2.1, computer use | 89.3% | 87.2% |
| Remote Labor Index, knowledge work | 41.3% at high effort | 21.3% |
| Humanity's Last Exam with tools | 74.2% | 67.7% |
| AA-Briefcase Elo, knowledge work | 1785 | 1800 |
The table reports a Fable 5.2 score and an Opus 5.5 score on twelve rows. Fable 5.2 leads on eleven of them. Only the Elo rating on AA-Briefcase goes the other way, with Opus 5.5 slightly ahead. The computer use gap is small, and the Remote Labor Index gap is large. The cybersecurity and protein rows compare Mythos 5.2 with Opus 5.5 run without safeguards, so we left them out.
Price per token or price per task
Per token, Fable 5.2 costs exactly twice Opus 5.5.
Per task, the gap can close or reverse, because a stronger model can finish in fewer turns and fewer tokens.
Anthropic's announcement gives one example. In early testing on FrontierCode 2.0, Fable 5.2 at high effort scored about level with Opus 5.5 at max effort, for roughly half the cost per task. That is one benchmark run by the vendor, and your workload may not behave the same way.
Effort moves cost more than the model choice does on many routes. Fable 5.2 defaults to high on the API, Opus 5.5 to medium, and Claude Code and the Claude apps use medium for both. A request that omits effort and moves from Opus 5.5 to Fable 5.2 therefore runs one level higher at twice the token price. Our Fable 5.2 pricing brief works through cache, batch and fast mode rates.
What changes in your code if you switch
Both models use the same Messages API, both reject forced tool choice and both run adaptive thinking that cannot be switched off. Anthropic's migration guide still lists differences that matter in production:
- Fable 5.2 needs 30-day data retention. Under zero data retention every request returns an error unless Anthropic has authorised it.
- Fable 5.2 runs the prefix check on every account, so code that edits earlier messages, the system prompt or the tool list fails there even if it works on Opus 5.5.
- Fable 5.2 has its own rate limit, with lower token limits at each tier.
- Opus 5.5 cannot read Fable 5.2's thinking blocks, so falling back from Fable to Opus drops the earlier reasoning.
- In organisations with customer-managed keys, structured outputs work on Opus 5.5 and not on Fable 5.2.
The full checklist, with the commands to run first, is in our Opus 5.5 to Fable 5.2 migration brief.
How to decide with your own numbers
- Take 50 to 100 real tasks from your logs, with the result you would have accepted for each.
- Run them on Opus 5.5 at medium and high effort, and on Fable 5.2 at medium and high, with the same prompts and tools.
- Score each run on task success, then record latency and total tokens, thinking included.
- Compare cost per successful task, not cost per request.
- Route only the task types where Fable 5.2 wins on that measure, and keep Opus 5.5 for the rest.
Where Digiton fits
Most teams we meet run one model for everything because nobody measured the alternative. Our production engineering review is built to give you a scorecard across ten layers with evidence for each, and model choice, effort and fallback are part of it. When the answer is a new agent or a rebuilt service, that work runs through our AI agent development and custom AI development services.
If you are choosing tools as well as models, our Claude Code vs Cursor guide and our Claude agent harness guide cover the layers around the model.
Frequently asked questions
Is Claude Fable 5.2 better than Opus 5.5?
On most of Anthropic's own benchmarks, yes: it leads on eleven of the twelve rows where both have a score. It also costs twice as much per token, and Anthropic still recommends Opus 5.5 as the starting point for most coding and knowledge work.
Is Fable 5.2 or Opus 5.5 better for coding?
Start with Opus 5.5 for everyday coding. Anthropic reports Fable 5.2 at 69.9% on FrontierSWE v2 against 62.3% for Opus 5.5, so move the hardest tasks, such as long migrations and multihour agent runs, to Fable 5.2 once your own tests show the gain.
How much more does Fable 5.2 cost than Opus 5.5?
Twice as much per token: $8 and $40 per million input and output tokens against $4 and $20. Cache reads cost $0.20 on both. Per task the gap depends on how many turns and tokens each model needs.
Can I switch from Opus 5.5 to Fable 5.2 by changing the model ID?
Often, yes. Check data retention, the prefix check, rate limits and structured outputs under customer-managed keys first, because each can turn a working Opus 5.5 request into an error on Fable 5.2.
What is the difference between Fable 5.2 and Mythos 5.2?
Anthropic says they are the same underlying model. Fable has stronger cybersecurity and biology safeguards and is available to everyone. Mythos 5.2 is for organisations verified through Anthropic's verification programs, at the same price.
Sources, checked 2026-11-09
6 sources
- Claude Fable 5.2 was released on 9 November 2026 with model ID claude-fable-5-2, a 1M token context window, 128K max output, $8 and $40 per million input and output tokens, default effort high and a June 2026 knowledge cutoff.Source: platform.claude.com, checked 2026-11-09.
- Claude Opus 5.5 was released on 22 September 2026 with model ID claude-opus-5-5, a 1M token context window, 128K max output, $4 and $20 per million input and output tokens and default effort medium.Source: platform.claude.com, checked 2026-11-09.
- Anthropic recommends starting most coding and knowledge tasks on Claude Opus 5.5 and moving up to Fable or Mythos depending on the scope and complexity of the work.Source: anthropic.com, checked 2026-11-09.
- In early testing on FrontierCode 2.0, Fable 5.2 at high effort scored about level with Opus 5.5 at max effort for roughly half the cost per task.Source: anthropic.com, checked 2026-11-09.
- Claude Fable 5.2 requires 30-day data retention, enforces the prefix check on every account and has a rate limit separate from Opus 5.5 with lower token limits at each tier.Source: platform.claude.com, checked 2026-11-09.
- Anthropic says to move to Claude Fable 5.2 when evals on a Claude Opus model at higher effort still do not meet your requirements.Source: platform.claude.com, checked 2026-11-09.
Digiton Dynamics OÜ is registered in Tallinn, Estonia, and delivers from Lisbon.
Update log
- Published. Specs, prices and benchmark rows copied from Anthropic's model pages, pricing page and announcement on the day of release.
Each source above was opened on the date beside it. How facts get chosen and checked is written down in the editorial policy.
Talk to an engineer about your system
Book a free 25-minute scoping call. We tell you what we would build, what it takes to keep it running, and whether we are the right fit.