Oleksandr Kuzmenko, EditorPublished 7 min readAI-assisted · editor-reviewed
TL;DR
- Fable 5.2 delivers 49.1% on FrontierCode 2.0 at Xhigh effort, outscoring prior models on code generation acceptance.
- API pricing is $8/M input and $40/M output tokens, with an optional 2.5x fast mode priced at $16/M input and $80/M output.
- Anthropic recommends using Fable 5.2 as an orchestrator or advisor alongside daily drivers like Opus 5.5 or Sonnet 5.5.
Key facts
- Input Token Price
- $8.00 / 1M tokens
- Output Token Price
- $40.00 / 1M tokens
- Fast Mode Pricing
- $16.00 in / $80.00 out per 1M tokens
- FrontierCode 2.0 (Xhigh)
- 49.1% (self-reported)
- FrontierSWE v2 Score
- 69.9% (self-reported)
- OSWorld 2.1 Score
- 89.3% (self-reported)
Source: Key facts — FrontierCode 2.0 (Xhigh), FrontierSWE v2 Score, OSWorld 2.1 Score
Source: Key facts. FrontierCode 2.0 (Xhigh): 49.1% (self-reported). FrontierSWE v2 Score: 69.9% (self-reported). OSWorld 2.1 Score: 89.3% (self-reported). The chart compares only values that share a unit. Each figure is copied from the story.
Metric chart
- 49.1% (self-reported)FrontierCode 2.0 (Xhigh)
- 69.9% (self-reported)FrontierSWE v2 Score
- 89.3% (self-reported)OSWorld 2.1 Score
- FrontierCode 2.0 (Xhigh)49.1% (self-reported)
- FrontierSWE v2 Score69.9% (self-reported)
- OSWorld 2.1 Score89.3% (self-reported)
Source: Key facts — Input Token Price, Output Token Price
Source: Key facts. Input Token Price: $8.00 / 1M tokens. Output Token Price: $40.00 / 1M tokens. The chart compares only values that share a unit. Each figure is copied from the story.
- $8.00 / 1M tokensInput Token Price
- $40.00 / 1M tokensOutput Token Price
Benchmark Highlights and Cost Reductions
Claude Fable 5.2 scores 49.1% on FrontierCode 2.0 at Xhigh effort, establishing a new benchmark record for code change merge acceptance in open-source repositories. On FrontierSWE v2, Fable 5.2 reaches 69.9% compared to 62.3% for Opus 5.5 and 65.5% for GPT-6 Astra. On AutomationBench, it hits 51.4%, while on Perplexity's WANDR search evaluation, it matches Opus 5.5's top performance at medium effort for roughly 40% of the token cost.
Production Migration and Architecture Roles
Anthropic recommends pairing models hierarchically rather than running Fable 5.2 for all interactions:
- Orchestrator Role: Fable 5.2 analyzes architecture, partitions broad tasks, and assigns subtasks to Claude Sonnet 5.5 or Haiku 5.5.
- Advisor Role: Faster daily drivers like Opus 5.5 consult Fable 5.2 exclusively when encountering unresolvable test failures or subtle concurrency bugs.
Internally, Anthropic leveraged an early iteration of Fable 5.2 to migrate an entire production service from Python to Rust, achieving a 26x improvement in startup speed.
Pricing and Fast Mode API Options
Fable 5.2 is available via the Anthropic API at $8 per million input tokens and $40 per million output tokens, reflecting a 30% speed boost over Fable 5.1. Developers needing low latency for real-time tool loops can enable fast mode at $16 per million input tokens and $80 per million output tokens, delivering up to 2.5x token emission speed.
Try it in 2 minutes
import anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-fable-5-2-2026",
max_tokens=4096,
messages=[
{"role": "user", "content": "Analyze architecture and propose Rust conversion for legacy Python service."}
],
extra_body={"effort": "high"}
)
print(response.content[0].text)When to use
- Use Fable 5.2 when orchestrating large-scale cross-repository refactors, complex language migrations, or deep security vulnerability audits.
- Use as an escalated reasoning step when smaller models fail automated integration test suites.
When NOT to use
- Do not use Fable 5.2 for standard boilerplate generation, single-file scripts, or routine daily pull requests where Opus 5.5 or Sonnet 5.5 are cheaper and sufficiently capable.
- Do not enable fast mode on batch pipelines where token generation latency does not affect user experience or agent loop execution.
What to do today
- Test Fable 5.2 on your most stubborn repository bugs or migrations by invoking it with high effort mode.
- Implement the Advisor pattern in agent frameworks to call Fable 5.2 only after Sonnet or Opus fails a verification step.
- Evaluate fast mode for latency-sensitive interactive coding pipelines requiring high reasoning depth.
Step names are taken from What to do today.
Step names are taken from What to do today. Test Fable 5.2 on your most stubborn repository bugs or migrations by invoking it with high effort mode.. Implement the Advisor pattern in agent frameworks to call Fable 5.2 only after Sonnet or Opus fails a verification step.. Evaluate fast mode for latency-sensitive interactive coding pipelines requiring high reasoning depth.
- Test Fable 5.2 on your most stubborn repository bugs or migrations by invoking it with high effort mode.
- Implement the Advisor pattern in agent frameworks to call Fable 5.2 only after Sonnet or Opus fails a verification step.
- Evaluate fast mode for latency-sensitive interactive coding pipelines requiring high reasoning depth.
What the community says
It'll be cheaper than hiring people and maintaining software, presumably
Really? A $200/mo plan gives you an enormous amount of inference on this model.