Sakana AI Unveils Fugu Max and Fugu Ultra v2, Pushing the Capability-Cost Frontier
On September 11, Sakana AI released two new versions of its multi-agent orchestration system Sakana Fugu: Fugu Max and Fugu Ultra v2. The company's core claim is that what the industry should really compete on isn't a single bigger model, but the "capability × cost" Pareto frontier—using dynamic orchestration to turn a static model menu into a schedulable system.
Confirmed
- Both products are built on the same multi-model orchestration architecture but are positioned differently: Fugu Max targets cost-effectiveness, priced at $2 input / $6 output per million tokens, with performance approaching top-tier models at 2-6x lower cost; Fugu Ultra v2 is the flagship.
- According to Sakana AI (relayed by hardmaru), Fugu Ultra v2 outperforms GPT-6 Astra and the Claude series on some coding benchmarks such as DeepSWE.
- Fugu Max extends the cost-effectiveness frontier by dynamically orchestrating open-weight models with specialized models; the company says it orchestrates the largest open-weight models to date.
Unconfirmed
- Claims like "surpassing GPT-6 Astra" and "2-6x lower cost" come from Sakana AI's official posts and reposts, with no independent third-party benchmarks yet.
Why it matters
- If orchestration strategies can truly approach flagship-model performance at significantly lower cost, it would upend the "stack parameters, compete on a single model" logic, offer new options for budget-conscious developers, and put pricing pressure on closed-source flagship models.
2026-09-11 ~ 2026-09-11 · 5 related posts
Primary sources
- [source] Sakana AI launches Fugu Max and Fugu Ultra v2, matching elite models at 2-6x lower cost — SakanaAILabs · 2026-09-11
- Sakana AI's Fugu Ultra v2 Claims to Beat GPT-6 Astra on Some Benchmarks; Fugu Max Cuts Output Price 40-60% — hardmaru · 2026-09-11
3 near-duplicate retellings: hardmaru · SakanaAILabs · SakanaAILabs