# Sakana Fugu

Sakana AI, released 2026-06-22

**Verdict: conditional.** Multi-agent orchestrator that matches frontier models on key benchmarks by routing across a pool of expert agents — but opaque routing and platform dependency limit transparency. Last verified 2026-06-23.

## Pricing

- Input: $5 / 1M tokens
- Output: $30 / 1M tokens
- Context: 272K tokens (>272K: $10/$45 per 1M)

## Strengths

- Frontier-competitive on SWE-Bench Pro (73.7), LiveCodeBench (93.2), GPQA-D (95.5) — matching Fable 5 and Mythos Preview without training a frontier model
- Multi-agent orchestration through a single OpenAI-compatible API — no SDK migration needed
- Agent pool is swappable — opt out specific models/providers for compliance
- Strong persona stability across long sessions — holds identity where other models drift
- Effective for code review — surfaces significantly more issues than single-model tools
- Opaque model routing — you cannot see which underlying models were used for a query
- Not available in EU/EEA at launch (GDPR compliance pending)
- Orchestration can consume 4-6× the tokens of a direct single-model call — hidden cost risk

## Benchmarks

- SWE-Bench Pro: 73.7 (Fugu Ultra) (https://sakana.ai/fugu/)
- TerminalBench 2.1: 82.1 (Fugu Ultra) (https://sakana.ai/fugu/)
- LiveCodeBench: 93.2 (Fugu Ultra) (https://sakana.ai/fugu/)
- LiveCodeBench Pro: 90.8 (Fugu Ultra) (https://sakana.ai/fugu/)
- Humanity's Last Exam: 50.0 (Fugu Ultra) (https://sakana.ai/fugu/)
- GPQA-D: 95.5 (Fugu Ultra) (https://sakana.ai/fugu/)
- CharXiv Reasoning: 86.6 (Fugu Ultra) (https://sakana.ai/fugu/)
- SciCode: 60.1 (Fugu) (https://sakana.ai/fugu/)

## Sources

- [Digital Applied — Fugu Analysis](https://www.digitalapplied.com/blog/sakana-fugu-multi-agent-orchestration-model-2026)
- [Sakana AI: Fugu pricing](https://sakana.ai/fugu/)
- [Startup Fortune — Fugu Launch](https://startupfortune.com/sakana-ai-launches-fugu-ultra-and-its-orchestration-model-matches-fable-and-mythos-without-training-a-single-frontier-model/)
- [VentureBeat — Fugu Launch Coverage](https://venturebeat.com/orchestration/no-claude-fable-5-no-problem-sakana-achieves-frontier-performance-with-new-fugu-multi-model-auto-synthesis-system)
- [Sakana Fugu Guide (Independent)](https://sakanafugu.com/)
- [ExplainX — Fugu Deep Dive](https://www.explainx.ai/blog/sakana-fugu-multi-agent-orchestration-model-2026)
- [Coursiv — Fugu vs Fable 5 Review](https://coursiv.io/blog/sakana-ai-fugu)

Canonical: https://neomanex.com/models/sakana-fugu