Sakana claims its AI model router Fugu Ultra v1.1 now beats Fable 5 without even including it in the pool
Quick Answer
Sakana AI's Fugu Ultra v1.1 claims to outperform Anthropic's Fable 5 by up to 7.9 points on benchmarks, despite Fable not being in its model pool.
Quick Take
The pricing remains at $5 per million input tokens and $30 per million output tokens, with a two-week training period for new models. Sakana still does not serve the EU or EEA due to regulatory concerns.
Key Points
- Fugu Ultra v1.1 shows a performance increase of up to 7.9 points over v1.0.
- The update includes a Claude Code-compatible endpoint for terminal access.
- Sakana's pricing remains at $5 per million input tokens and $30 per million output tokens.
- Fugu has been available on platforms like OpenRouter and Vercel since launch.
- Sakana does not serve the EU or EEA due to GDPR and regulatory issues.
📖 Reader Mode
~1 min readSakana AI has released Fugu Ultra v1.1, an update to the AI router that distributes each query across a pool of publicly available top-tier models. The company claims performance gains of up to 7.9 points over v1.0, with the biggest jumps on ProgramBench and TerminalBench 2.1. Fugu v1.1 reportedly beats Fable 5, even though Fable 5 isn't part of the router's selection pool. All of these numbers come from Sakana itself, and no independent verification exists yet.
Pricing stays at $5 per million input tokens and $30 per million output tokens. Sakana says it takes about two weeks of training and evaluation before a new top-tier model gets added to the pool. The architecture is described in the technical report. The update also adds a Claude Code-compatible endpoint for calling Fugu directly from the terminal. Since launch, Fugu has been available on platforms like OpenRouter and Vercel.
The first Fugu version got a lukewarm reception. Critics pointed to high token usage, slow speed, and poor results. Sakana still doesn't serve the EU or EEA, citing GDPR and "EU-specific regulations."
— Originally published at the-decoder.com
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from The Decoder
See more →
An AI model programmed nonstop for 19 days on a single MirrorCode task that cost $2,600 to run
Epoch AI's MirrorCode benchmark reveals Claude Opus 4.7 as the leader with a 56% solve rate, reconstructing a 16,000-line toolkit in 14 hours. Despite this, all models tested struggle with the most complex tasks, highlighting limitations in current AI capabilities. The single task consumed $2,600 over 19 days, raising questions about cost-effectiveness in AI development.
