Back to feed
News Story
THE DECODER
1 sources

Sakana claims its AI model router Fugu Ultra v1.1 now beats Fable 5 without even including it in the pool

Sakana AI released Fugu Ultra v1.1, an AI model router that claims up to 7.9-point gains over v1.0 and outperforms Fable 5 without including it in the routing pool. The update adds a Claude Code-compatible endpoint but remains unavailable in the EU. Independent verification is pending.

SynthePulse Insight · AI deep reading

Fugu Ultra v1.1: Sakana Claims Routing Model Beats Unpooled Fable 5

Version 1 · 1 source

Sakana AI releases Fugu Ultra v1.1, claiming performance gains of up to 7.9 points and beating Anthropic's Fable 5 on most benchmarks—even though Fable 5 is not in the router's model pool. All data comes from Sakana itself, with no independent verification.

  • Sakana AI releases Fugu Ultra v1.1, claiming up to 7.9-point performance improvement over v1.0, with largest gains in ProgramBench and TerminalBench 2.1.
  • Sakana claims Fugu v1.1 beats Anthropic's Fable 5 on most benchmarks, but Fable 5 is not an optional model in the router.
  • Pricing remains at $5 per million input tokens and $30 per million output tokens.
  • New Claude Code-compatible endpoint added, allowing direct terminal calls to Fugu.
  • Service still unavailable in EU/EEA, citing GDPR and 'EU-specific regulations'.
  • All performance claims come from Sakana itself, with no independent verification.
Open section navigationUpdate Details and Performance Claims

Update Details and Performance Claims

Sakana AI released Fugu Ultra v1.1 on July 24, 2026, an updated version of its AI model router. The router assigns each query to a pool of publicly available top-tier models. Sakana claims v1.1 achieves up to 7.9-point performance improvement over v1.0, with the largest gains in ProgramBench and TerminalBench 2.1.

Sakana further claims Fugu v1.1 beats Anthropic's Fable 5 on most benchmarks, even though Fable 5 is not included in the router's model selection pool. All performance data comes from Sakana itself, with no independent verification yet.

Pricing and Availability

Pricing remains unchanged: $5 per million input tokens and $30 per million output tokens. Sakana states that adding a new top-tier model to the pool takes about two weeks of training and evaluation.

The update also adds a Claude Code-compatible endpoint, allowing users to call Fugu directly from the terminal. Since its initial release, Fugu has been available on platforms like OpenRouter and Vercel.

However, the service remains unavailable in the EU or EEA, which Sakana attributes to GDPR and 'EU-specific regulations.'

Early Version Reception and Current Uncertainty

Reception of the first Fugu version was lukewarm, with critics pointing to high token usage, slow speed, and poor results. Whether the current v1.1 version addresses these issues is unclear, as all improvement claims come from Sakana itself without independent verification.

Sakana's claim that Fugu v1.1 beats the unpooled Fable 5 is particularly striking, but likewise lacks third-party confirmation.

Credibility boundary

All performance claims in this article come from Sakana AI itself, with no independent verification. Pricing, availability restrictions, and architecture descriptions come from Sakana's official release, but critical feedback comes from early versions.

Insight takeaway

Fugu Ultra v1.1 claims significant performance gains and beats the unpooled Fable 5, but all claims require independent verification. Pricing remains unchanged, and EU users still cannot access the service.

Primary report

THE DECODER

Primary source