Sakana AI says its Fugu Ultra v1.1 update raises the bar for an AI model router: a system that decides which model should handle each query from a pool of publicly available top-tier models. The headline claim is direct: Fugu v1.1 reportedly beats Fable 5, even though Fable 5 is not one of the models available for the router to select.
The update is still best read with caution. The performance numbers come from Sakana itself, and the source notes that no independent verification exists yet.
What Fugu Ultra v1.1 Changes
Fugu Ultra v1.1 is an update to Sakana AI's router rather than a single model presented in isolation. Its role is to distribute each incoming query across a selection pool of publicly available top-tier models.
That architecture matters because the router is being judged by the combined result of its decisions. Instead of asking whether one fixed model is strongest in every situation, the approach asks whether routing can produce better outcomes by choosing among available options.
Sakana says the new version delivers performance gains of up to 7.9 points over v1.0. The largest reported improvements are on ProgramBench and TerminalBench 2.1.
The most attention-grabbing part of the release is the comparison with Fable 5. According to Sakana, Fugu v1.1 beats Fable 5 even though Fable 5 is not included in the router's selection pool.
Why The Fable 5 Claim Stands Out
For an AI model router, beating a model that is not in its own pool is a specific kind of claim. It suggests the router's choices among other publicly available top-tier models can, in Sakana's reported testing, produce stronger aggregate performance than Fable 5.
That does not mean Fable 5 was used by Fugu v1.1. The source is explicit that Fable 5 is not part of the router's selection pool. The point of the claim is that Fugu v1.1 can outperform it without directly calling it.
There are two important limits to keep in view:
- The reported benchmark numbers come from Sakana itself.
- No independent verification exists yet.
Those limits do not make the claim irrelevant. They do mean readers should separate the product announcement from confirmed outside validation. For teams tracking AI routing, the claim is notable, but the evidence described so far is still company-provided.
Pricing And Availability Stay Familiar
Sakana is keeping pricing unchanged with this update. Fugu Ultra v1.1 stays at $5 per million input tokens and $30 per million output tokens.
The company also says that adding a new top-tier model to the router's pool takes about two weeks of training and evaluation. That timing gives a sense of how the system is maintained: new models are not simply dropped in instantly, at least according to Sakana's description.
Since launch, Fugu has been available on platforms like OpenRouter and Vercel. The v1.1 update also adds a Claude Code-compatible endpoint, allowing Fugu to be called directly from the terminal.
That terminal access is practical for developers who want to use Fugu in a command-line workflow. It also places the router closer to coding and evaluation environments where tools such as ProgramBench and TerminalBench 2.1 are especially relevant.
The Earlier Reception Still Matters
The first version of Fugu did not arrive to universal enthusiasm. The source describes its reception as lukewarm, with critics pointing to high token usage, slow speed, and poor results.
That background gives Fugu Ultra v1.1 more to prove. The update is not only about posting higher benchmark numbers. It is also about answering earlier concerns around efficiency, latency, and quality.
The source does not say whether the new version directly resolves those criticisms. It only reports Sakana's performance claims, pricing, model onboarding timeline, terminal endpoint, and availability context. Without independent testing, the practical effect of the update remains an open question.
Where Access Is Still Limited
Fugu Ultra v1.1 is not available everywhere. Sakana still does not serve the EU or EEA.
The reason given is regulatory. Sakana cites GDPR and "EU-specific regulations" as the basis for that limitation.
For users and organizations in those regions, the benchmark claims and new endpoint may be interesting but not immediately usable through Sakana's service. For users outside those regions, the update presents a clearer proposition: a router with unchanged pricing, a new terminal path, and a company-reported benchmark jump over v1.0.
The central takeaway is simple. Sakana is positioning Fugu Ultra v1.1 as a stronger AI model router, not by adding Fable 5 to its pool, but by claiming better results without it. The next step for the claim is independent verification.