Claude Fable 5.2 · released November 9, 2026
Hire engineers who ship with Claude Fable 5.2
Anthropic released Claude Fable 5.2 today. Your competitors got the same API key. The teams that get value from it next quarter will be the ones with evals, agents and guardrails built around it. We place engineers who do that work, inside your repo, in 2 to 4 weeks, from $3K a month.
API model ID
claude-fable-5-2
Price
$8 input / $40 output per million tokens
Fast mode
Up to 2.5x faster output at $16 / $80
Context window
1M tokens, 128K max output
Default effort
high on the API, medium in Claude Code
Retirement
Not sooner than November 9, 2027
Sources: Anthropic announcement and Claude models overview. Compare it with other models on our model hub.
Why the model alone doesn't ship product
Switching the model ID is a one-line change. Then four items from Anthropic's own release notes land on an engineer's desk.
01
Effort is a cost decision, and the defaults differ
Fable 5.2 defaults to high effort on the API and medium in Claude apps and Claude Code. Anthropic's own numbers show the spread matters: on FrontierCode 2.0, Fable 5.2 at high effort lands about level with Claude Opus 5.5's best score at roughly half the cost per task. Someone has to pick the effort level per job, measure it and keep the bill in check.
02
It's the expensive tier, so routing matters
At $8 in and $40 out per million tokens, Fable 5.2 costs twice Opus 5.5 per token. Anthropic's docs tell teams to start on Opus 5.5 and move up to Fable 5.2 when evals at higher effort still fall short. Without evals, you can't tell which calls need Fable and which are burning money.
03
Safeguards change what the model will do
Fable 5.2 blocks high-risk cyber work such as penetration testing and exploit generation, plus most research biology tasks. Blocked cyber queries fall back to Opus 4.8 and blocked biology queries to Opus 5. If your product touches security tooling, an engineer has to design for that fallback before users find it.
04
Integrations need a migration pass
Anthropic says developers should update integrations for preserved thinking, which now applies to all API accounts. Fast mode runs only on some platforms: Claude Code, the Claude Platform, Claude Models on Google Cloud and Microsoft Foundry. The models overview doesn't list an Amazon Bedrock ID for Fable 5.2 yet. Each of those is a config change somebody owns.
What a Fable 5.2 engineer actually does
The job maps onto what Anthropic says the model is built for. Here is the work on a client team.
Agents that run for hours
Anthropic positions Fable 5.2 for long-horizon agentic work and describes two patterns: Fable as an Advisor that Haiku, Sonnet or Opus consult, and Fable as an Orchestrator that hands simpler tasks to smaller models. Uber ran a 10 to 12 hour security audit with about 170 reviewer agents. The engineer designs that split, the tools each agent can call and where a human approves.
Claude Code workflows on your repo
Anthropic says over 90% of changes merged into its core codebase are written by Claude with human guidance. On FrontierSWE v2, Fable 5.2 scores 69.9% against 56.3% for Fable 5.1. Getting that inside your team means CLAUDE.md files, hooks, subagents, review gates and CI checks set up for your codebase.
Evals before and after the switch
A model upgrade is a regression risk until you measure it. The engineer builds a test set from your real tickets and documents, scores Opus 5.5 and Fable 5.2 at different effort levels, and writes down cost per task next to the pass rate. That table is what decides the rollout.
Guardrails and cost controls
Tool permissions, sandboxed execution, spend caps, prompt caching (cache reads cost 2.5% of the base input price on Fable 5.2), Batch API for jobs that can wait (50% off), and logging you can audit. Anthropic's report shows sandbox escape attempts dropped when testers described the sandbox and added tools for reporting weaknesses. The engineer makes those calls.
Want the Claude Code side handled as a project instead of a hire? See our Claude Code development agency.
How to vet a Claude Fable 5.2 engineer
Fable 5.2 is hours old, so nobody has years on it. Hire for the habits that carry over: evals, cost control and agent design. Use these six questions.
1. Walk me through the last agent you shipped. What eval did you run, and what was the final number?
Real builders answer with a metric and a test set. Weak candidates describe a demo.
2. Fable 5.2 defaults to high effort on the API and medium in Claude Code. How would you pick effort for a support-ticket agent?
You want to hear: measure pass rate and cost per task at two or three levels, then pick the cheapest one that clears the bar.
3. When would you keep a workload on Opus 5.5 or Sonnet 5.5 instead of moving it to Fable 5.2?
Tests whether they route by evidence. Anthropic's own guidance is to start on Opus 5.5 and move up when evals fall short.
4. A Fable 5.2 call in our security tooling falls back to Opus 4.8. How do you detect it and what do you change?
Checks they read the release notes and think about safeguards as product behaviour.
5. Show me a CLAUDE.md or hook config you wrote and explain one rule you added after something broke.
Claude Code fluency shows up in rules someone wrote after an incident.
6. How do you stop a long-running agent from spending $2,000 overnight?
Look for spend caps, step limits, caching, batch jobs for non-urgent work and alerts.
A take-home that takes four hours
Give the candidate 30 real (anonymized) support tickets and a read-only tool that looks up order status. Ask for an agent that drafts replies, plus:
- an eval script that scores the drafts against 10 tickets you graded by hand
- a table of pass rate and cost per ticket for Opus 5.5 and Fable 5.2 at two effort levels
- one guardrail that stops the agent from promising a refund
- a README that says which model they would ship, and why
Read the README first. A candidate who picks the cheaper model because the numbers say so is the one you want. Pay for the time. More screening questions are in our guide to hiring Claude Code developers.
Claude Fable 5.2 for startups
Fable 5.2 launched cheaper than Fable 5.1, which cost $10 in and $50 out per million tokens. It is still twice the per-token price of Opus 5.5. For a startup with one or two engineers, that gap decides whether the model is worth it, so the rules we give placed engineers are simple:
- Default to Opus 5.5. Move a workload to Fable 5.2 only when an eval on your own data shows Opus 5.5 at higher effort falls short. That is Anthropic's own advice.
- Pay less for repeat context. Cache reads on Fable 5.2 cost 2.5% of the base input price, and the Batch API takes 50% off for jobs that can wait.
- Use fast mode on purpose. It is up to 2.5x faster and costs double. Keep it for interactive work where someone is waiting.
- Treat speed claims as claims. Anthropic says Fable 5.2 is up to 30% faster than Fable 5.1. Independent speed and price rows from Artificial Analysis are not published yet, so time it on your own prompts.
A startup usually doesn't need a full-time Fable specialist. It needs one engineer who can run these checks and ship the agent. That is the role we place from $3K a month.
Or skip the hiring loop
We run those checks before you meet anyone. Our engineers already build agents with Claude Code, MCP and sandboxed execution for client teams, and they move to Fable 5.2 where your evals say it pays off.
2-4 weeks
from request to start date
From $3K/mo
no recruitment fee, no employer taxes on top
90 days
replacement guarantee if the fit is wrong
See what placed engineers have shipped in our case studies, or read the full offer on hire AI developers.
FAQ
What is Claude Fable 5.2?
Claude Fable 5.2 is Anthropic's model for demanding reasoning and long-horizon agentic work, released on November 9, 2026. The API model ID is claude-fable-5-2. It costs $8 per million input tokens and $40 per million output tokens, with a 1M token context window and 128K max output.
Do I need a Fable 5.2 specialist, or can my current engineers pick it up?
Strong engineers can learn the API in a day. The slower part is the system around it: evals, effort and cost tuning, agent orchestration, guardrails and the migration for preserved thinking. If nobody on your team has shipped an agent to production, a placed engineer who has done it saves you the learning curve.
How fast can AY Automate place a Claude Fable 5.2 engineer?
Placement usually takes 2 to 4 weeks from request to start. Engineers start from $3K a month with no recruitment fee, and every placement carries a 90-day replacement guarantee.
What does the 90-day replacement guarantee cover?
If the engineer is not the right fit in the first 90 days, we replace them at no extra cost.
Is Fable 5.2 worth the price over Claude Opus 5.5?
Only where your evals say so. Fable 5.2 costs twice Opus 5.5 per token, and Anthropic recommends starting on Opus 5.5 and moving up when evals at higher effort fall short. Anthropic also reports that Fable 5.2 at high effort roughly matches Opus 5.5's best FrontierCode 2.0 score at about half the cost per task, so the answer depends on the workload.
Is Claude Fable 5.2 worth it for startups?
For most startup workloads, start on Claude Opus 5.5, which costs half as much per token, and move a task to Fable 5.2 when your evals show Opus 5.5 falls short. Fable 5.2 is cheaper than Fable 5.1 ($8 and $40 per million tokens, down from $10 and $50), and prompt caching and the Batch API cut the bill further.
What about Claude Mythos 5.2?
Anthropic says Mythos 5.2 is the same underlying model as Fable 5.2 without the cyber and biology safeguards. Access requires the Cyber Verification Program or the Life Sciences Verification Program. Most product teams will build on Fable 5.2.
Put a Fable 5.2 engineer on your backlog
Tell us what you want to build and which models you run today. On a free call we'll say whether Fable 5.2 pays off for it, what the role looks like and who on our team fits.
Book a free strategy call