Rakuten Speeds Up Agent Workflows 10x Using Claude Fable 5

Anthropic published a case study detailing how Rakuten Group transitioned from basic code assistance to deploying long-running, autonomous agents using Claude Fable 5. The implementation allows Rakuten to execute multi-step workflows overnight with minimal human intervention.

Rakuten utilizes managed agents that operate for hours or days to refactor large codebases and conduct deep research. This shift represents a move from single-prompt completion to asynchronous task resolution.

Performance Metrics

Rakuten measures its internal progress on a proprietary Rakuten-SWE-Bench. Fable 5 resolved three times more production tasks than previous models in this environment. Yusuke Kaji, General Manager of AI for Business at Rakuten, reported a 10x acceleration in task processing for workflows that typically require days of manual developer effort.

The results highlight the impact of self-reflection capabilities when you evaluate and test AI agents in production environments.

Claude Fable 5 Architecture

Released in June 2026, Claude Fable 5 represents Anthropic’s “Mythos-class” intelligence tier. The model features a 1-million-token context window and a 128K maximum output limit.

Its primary differentiator is a built-in verification loop. The model automatically writes tests and validates its own logic before finalizing outputs. This architectural decision directly impacts how it compares to alternatives in the GPT vs Claude vs Gemini landscape.

Model	SWE-bench Pro Score
Claude Fable 5	80.3%
GPT-5.5	58.6%
Gemini 3.1 Pro	54.2%

Pricing and Subscription Access

Anthropic prices Fable 5 higher than its standard Opus tier, reflecting the increased compute required for the verification loop.

Token Type	Cost per Million Tokens
Input	$10.00
Output	$50.00

Following a temporary export control suspension in June, Anthropic shifted its access model. Starting July 20, Fable 5 is permanently included in specific subscription tiers rather than operating strictly on usage credits. Max and Team Premium plans receive access at 50% of weekly usage limits. Pro and Team Standard users can access the model via usage credits, with existing users receiving a one-time $100 allocation.

If you build multi-agent systems for complex enterprise tasks, the transition of Fable 5 to stable subscription plans changes your compute forecasting. You can now rely on a predictable cost structure when deploying agents that require extensive self-correction and validation over long running sessions.