Sakana AI’s Fugu Ultra v1.1 Outperforms Fable 5 in Latest Update

Sakana AI has released a significant update to its intelligent model router, Fugu Ultra v1.1, claiming massive performance leaps across critical technical benchmarks. The update marks a strategic attempt to refine how queries are distributed across top-tier models to achieve superior reasoning and coding capabilities.

Breaking Benchmarks: The Fugu Ultra v1.1 Advantage

The core innovation of Fugu Ultra v1.1 lies in its routing intelligence, which selects the most suitable model from a pool of publicly available LLMs for any given prompt. Sakana AI reports that this version delivers performance gains of up to 7.9 points over the initial v1.0 release.

Most notably, the router demonstrated significant jumps in specialized technical assessments, specifically on the ProgramBench and TerminalBench 2.1 benchmarks. In a striking claim, Sakana asserts that Fugu Ultra v1.1 now outperforms Anthropic’s Fable 5 across most benchmarks—even though Fable 5 is not actually included in the router's selection pool. While these results currently lack independent third-party verification, they suggest that the routing logic is becoming highly efficient at extracting peak performance from existing model architectures.

Technical Architecture and Developer Integration

Sakana’s approach to maintaining a cutting-edge pool is rigorous; the company states it requires approximately two weeks of dedicated training and evaluation before any new top-tier model is officially integrated into the Fugu ecosystem. This ensures that the router's decision-making process remains optimized for the latest weights and capabilities in the market.

For developers, the update introduces a new Claude Code-compatible endpoint, allowing users to call Fugu directly from their terminal. This integration streamlines workflows for engineers who require high-level reasoning and terminal-based automation. Despite these technical strides, the pricing remains consistent with the previous version: $5 per million input tokens and $30 per million output tokens. Fugu is currently accessible through major platforms including OpenRouter and Vercel.

Addressing Market Challenges and Regulation

The release comes after a period of scrutiny for the initial Fugu launch. Early critics highlighted issues regarding high token consumption, latency, and inconsistent results. With v1.1, Sakana appears to be doubling down on efficiency and specialized task performance to win back developer trust.

However, geographical limitations remain a hurdle for global adoption. Sakana AI currently does not serve users within the European Union (EU) or the European Economic Area (EEA), citing complexities surrounding GDPR and other EU-specific regulatory frameworks. As the "model router" category grows, Sakana’s ability to prove these performance claims through transparent, verifiable data will be the ultimate test of its viability in the competitive AI landscape.

Key Takeaways

  • Superior Benchmarking: Fugu Ultra v1.1 shows a 7.9-point improvement over v1.0, notably outperforming Anthropic's Fable 5 on several metrics despite Fable 5 not being in the router's pool.
  • Developer-Centric Updates: The new version adds a Claude Code-compatible terminal endpoint, facilitating easier integration for coding workflows.
  • Rigorous Curation: Sakana employs a two-week evaluation cycle for every new model added to the router to maintain high accuracy and performance.