Fable 5.2 vs Opus 5.5 is Anthropic's own top-two decision: Fable 5.2, released on 9 November 2026, scores higher on almost every benchmark Anthropic ran, and Opus 5.5 costs half as much per token and is the model Anthropic tells most people to start with.
Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →
For coding agents the gap is real but narrower than the headline suggests, and effort settings change the maths completely.
If you want models like these working on your SEO and content, get your Backlink Gameplan, or try Polar Rank free for 3 days.
Everything below comes from Anthropic's models overview, the two model pages and the Fable 5.2 launch post.
Spec sheet: claude-fable-5-2 vs claude-opus-5-5
| Spec | Claude Fable 5.2 | Claude Opus 5.5 |
|---|---|---|
| Released | 9 November 2026 | 22 September 2026 |
| Anthropic's description | Demanding reasoning and long-horizon agentic work | Long-running agentic coding and knowledge work |
| Default effort | high | medium |
| Thinking | Adaptive, always on | Adaptive, always on |
| Context / max output | 1M / 128K tokens | 1M / 128K tokens (300K on Batches with beta header) |
| Knowledge cutoff | June 2026 | June 2026 |
| Relative list price per token | 2x Opus 5.5 | Baseline |
| Cache read rate | 2.5% of base input | 5% of base input |
| Data retention | 30 days required; no ZDR unless authorised | Standard |
| Retirement floor | Not before 9 Nov 2027 | Not before 22 Sep 2027 |
Notice the default effort row.
A request that omits effort runs Fable 5.2 at high and Opus 5.5 at medium, so an untuned side-by-side test compares two different settings and makes Fable look slower and pricier than it needs to be.
Coding benchmarks Anthropic published
| Benchmark | Fable 5.2 | Opus 5.5 | Note |
|---|---|---|---|
| FrontierCode 2.0 (would maintainers merge it?) | 49.1% | 45.5% | Fable at Xhigh; best released score Anthropic tested |
| FrontierCode 2.0 at High effort | 45.3% | 45.5% (best) | Roughly half the cost per task for Fable |
| FrontierSWE v2 | 69.9% | 62.3% | Agentic coding |
| OSWorld 2.1 (offline, partial) | 89.3% | 87.2% | Computer use |
| AutomationBench | 51.4% | 42.5% | Business workflows |
| BenchCAD 2.1 (Max effort) | 83.5 | 65.3 | Hardware design, 0-100 |
The row that matters most for budgets is the second one.
At High effort Fable 5.2 matched Opus 5.5's best FrontierCode score for roughly half the cost per task, according to Anthropic, because it reaches the same quality with fewer tokens.
So "twice the price per token" does not mean twice the cost per finished task.
Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →
Cost per token vs cost per task
Anthropic's pricing page lists Fable 5.2 at double Opus 5.5's per-token rate for input and output.
Batch processing halves both, and cache reads on Fable 5.2 are billed at 2.5% of its base input rate against 5% on Opus 5.5.
Fable 5.2 also costs less per token than Fable 5.1 did.
Then there is effort: Anthropic's charts show Fable 5.2 at Medium effort matching Opus 5.5's top WANDR search score for about 40% of the cost, and at Low effort matching Opus 5.5's best Humanity's Last Exam score for less than a tenth.
The cheapest setup is often Fable 5.2 at a lower effort, not Opus 5.5 at its maximum.
🔥 Running Fable 5.2 and Opus 5.5 in one Claude Code setup? Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members testing the newest models on real business work. → Join the AI Profit Boardroom
Effort settings that change the result
Both models use the effort parameter with adaptive thinking always on; disabling thinking or sending budget_tokens returns a 400 error.
import anthropic
client = anthropic.Anthropic()
# Opus 5.5 at its default for routine work
client.messages.create(model="claude-opus-5-5", max_tokens=32000,
output_config={"effort": "medium"},
messages=[{"role": "user", "content": "Fix the failing test in billing/"}])
# Fable 5.2 at high for the stubborn migration
client.messages.create(model="claude-fable-5-2", max_tokens=64000,
output_config={"effort": "high"},
messages=[{"role": "user", "content": "Plan and carry out the ORM migration"}])
Anthropic's Fable 5.2 notes add three behaviours to plan for.
A prompt line asking for less thinking does not stop Fable 5.2 thinking at medium and above.
On long multistep tasks it sometimes ends a turn with a list of what remains, more often at low and medium.
And it delegates readily to subagents when your harness invites it, which can drive a large share of a session's cost.
Tool use and API rules
- Forced tool use,
tool_choiceset toanyor a named tool, returns a 400 on both models; keepautoand use strict tool use for schema-valid JSON. - On Fable 5.2, omit
temperature,top_pandtop_k. - Fable 5.2 thinking blocks can only be read by Fable 5.2 or Mythos 5.2, so mixed-model conversations need care.
- A Fable 5.2 executor needs a Fable 5.2 or Mythos 5.2 advisor if you use the advisor pattern.
- Fable 5.2 uses the computer use toolset rather than
computer_20251124.
Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →
Speed and fast mode
Anthropic says Fable 5.2 generates output up to 30% faster than Fable 5.1, and both Fable 5.2 and Opus 5.5 are listed at moderate latency.
Fable 5.2 also has an optional fast mode that speeds up token generation by up to 2.5 times at premium pricing, available in Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry.
For interactive coding where you sit and wait, fast mode can be the difference between Fable feeling sluggish and feeling usable.
📞 Want Claude working on your search traffic? Get your Backlink Gameplan and we will map where Opus and Fable fit into your SEO and AI search work, or try Polar Rank free for 3 days. Backlink Gameplan or try Polar Rank free for 3 days
What Anthropic recommends
Anthropic's guidance is explicit: start most coding and knowledge tasks on Opus 5.5, its strongest daily driver, and move up to Fable 5.2 when the scope or complexity calls for it.
Fable 5.2 can also act as an Advisor that Haiku, Sonnet or Opus models consult on hard problems, or as an Orchestrator that hands simpler tasks to smaller models.
That orchestrator pattern is the one most agent builders should try first: Fable plans and reviews, Opus or Sonnet does the volume.
Coding evidence beyond benchmarks
Anthropic says more than 90% of the changes merged into its core codebase are now written by Claude with human guidance.
It reports a team of agents on an earlier Fable 5.2 build migrating a production service from Python to Rust, which now starts about 26 times faster.
And in one experiment Fable 5.2 coordinated over 1,000 subagents for eight days to build a prototype operating system that boots on a laptop.
Those are Anthropic's own accounts, but they show the kind of long, unattended job Fable 5.2 is aimed at.
Which to run
| Your workload | Pick | Setting |
|---|---|---|
| Daily feature work and bug fixes | Opus 5.5 | medium |
| Large refactors and migrations | Fable 5.2 | high |
| Overnight unattended agents | Fable 5.2 | high or xhigh |
| Quick reviews and test fixes | Opus 5.5 | low or medium |
| Planner over cheaper workers | Fable 5.2 | high, as Orchestrator |
| Security tooling | Neither by default | Fable 5.2 safeguards block most cyber tasks |
Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →
Migration checklist
- Set
effortexplicitly on both models; do not rely on defaults. - Remove
budget_tokens, disabled thinking and forcedtool_choice. - On Fable 5.2, remove sampling parameters.
- Check whether your account allows 30-day retention, which Fable 5.2 requires.
- Configure refusal fallback so a declined request retries on another model.
- Measure cost per finished task over a week, not cost per call.
Defaults inside Claude Code
The API default and the app default are not the same thing.
Anthropic says Medium effort is the default for Fable 5.2 in the Claude apps and in Claude Code, while a raw API request without effort runs at high.
If your benchmark numbers come from the API but your developers live in Claude Code, you are measuring two different set-ups.
Pin the effort in both places, or at least write down which one each team is using.
Caching and batch for long agent loops
Long agent sessions resend the same system prompt, tool definitions and repository context on every turn, so prompt caching dominates the bill.
Fable 5.2 bills cache reads at 2.5% of its base input rate, half the 5% share Opus 5.5 pays, which narrows the effective price gap on cache-heavy loops.
For jobs that can wait, the Message Batches API cuts both models to half price, which suits nightly code review, test generation and documentation sweeps.
Stack the two and an overnight Fable 5.2 batch with a warm cache can land much closer to Opus 5.5 than the list price suggests.
Subagent cost control
Fable 5.2 delegates to subagents readily, and Anthropic warns those subagents can drive a large share of a session's total spend.
Give the orchestrator a cap on concurrent subagents, route subagents to Opus 5.5 or a Sonnet model by default, and log tokens per subagent so you can see where the money goes.
Without those limits, a single ambitious migration can fan out further than you planned.
Refusals and fallback
Fable 5.2 carries cyber and biology safeguards that block most cybersecurity tasks, including some that look routine to a developer, such as writing a port scanner for your own network.
Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →
Anthropic's refusal and fallback guide shows how to retry a declined request on another model automatically.
For coding agents, set Opus 5.5 as the fallback so a refused step does not stall the whole run, and log every fallback so you can spot patterns.
Bottom line for operators
Opus 5.5 stays the default for most daily coding; Fable 5.2 is the model you promote the hard, long and ambiguous jobs to.
Because Fable 5.2 reaches Opus-level quality at lower effort, test it at medium before you assume it is too expensive.
Members of the AI Profit Boardroom get the Agent OS and Claude Code workflows built around both, and you can get your Backlink Gameplan for the SEO side, or try Polar Rank free for 3 days.
Get Your Backlink Gameplan →Try Polar Rank Free for 3 Days →
FAQ: Fable 5.2 vs Opus 5.5
Is Fable 5.2 better than Opus 5.5 for coding?
On Anthropic's tests, yes: 49.1% vs 45.5% on FrontierCode 2.0 at top settings and 69.9% vs 62.3% on FrontierSWE v2. Anthropic still recommends starting most coding on Opus 5.5.
How much more does Fable 5.2 cost?
Its list price per token is double Opus 5.5's. At High effort it matched Opus 5.5's best FrontierCode score for roughly half the cost per task, so cost per job can be lower.
What is the default effort on each?
Fable 5.2 defaults to high and Opus 5.5 to medium on the Claude API.
Does Fable 5.2 have fast mode?
Yes. Fast mode speeds up token generation by up to 2.5 times at premium pricing, including in Claude Code.
Can I turn thinking off on either model?
No. Adaptive thinking is always on for both; control depth with the effort parameter.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Set effort explicitly and measure cost per finished task, and Fable 5.2 vs Opus 5.5 settles itself; for SEO help, get your Backlink Gameplan, or try Polar Rank free for 3 days.











