Sakana Fugu Ultra is for teams that need frontier-level reasoning and coding quality but cannot depend on a single US-based model provider like Claude Opus 4.8. It replaces a direct single-model subscription with a swappable multi-model backend, trading predictable response times for coordinated multi-agent accuracy across a 128,000-token max output.
Sakana Fugu is an orchestration model from Sakana AI that scored 95.5% on GPQA Diamond in its June 2026 technical report, the highest vendor-reported score among frontier models at launch. Rather than a single dense model, it routes each query across a pool of frontier LLMs using its TRINITY and Conductor coordination frameworks, then returns one synthesized answer.
Where it sits
- $11.25/M$ per 1M tokensBlended price (3:1)Lower is better#55 / 64peer median $1.70/Mvendor price, checked by HokAI
- --tokens/sOutput speedHigher is better-- / 39peer median 90 tok/scited: Artificial Analysis
- --% solvedSWE-bench VerifiedHigher is better-- / 28peer median 78.3%per source, see benchmark scores
- 95.5%% correctGPQA DiamondHigher is better#2 / 44peer median 88.3%per source, see benchmark scores
Pricier than 90% of the 64 GA models with a published price, in the top third on GPQA Diamond (rank 2 of 44), and one of 65 whose vendor states it does not train on customer data.
Ranks are against GA models on HokAI that publish the same figure; ties share a rank.
Provider: Sakana AI · Family: Sakana Fugu
Context window: 1,000,000 tokens · Max output: 128,000
Input modalities: text, tool-calls · Output: text, tool-calls
About Sakana Fugu
Sakana Fugu is a multi-agent orchestration model released on June 22, 2026 by Sakana AI, a Tokyo-based startup. Unlike a standard single-model LLM, Fugu is itself a language model trained to coordinate a pool of other frontier LLMs. When you send a request to the Fugu API, Fugu decides whether to answer directly or to break the problem into subtasks, delegate those subtasks to a team of specialist models, verify the outputs, and synthesize one final response. The product exposes two tiers: Fugu (optimized for lower latency on everyday tasks) and Fugu Ultra (tuned for maximum accuracy on hard, multi-step problems with a 1,000,000-token extended context window).
The architecture is grounded in two research frameworks Sakana published at ICLR 2026. TRINITY uses a lightweight evolved coordinator that assigns roles to each model in the pool: Thinker (plans the approach), Worker (executes subtasks), and Verifier (checks outputs). The coordinator learns how to allocate a limited turn budget across these roles rather than following a fixed workflow. Conductor is a complementary framework that learns natural-language coordination strategies via reinforcement learning, generating custom instructions for how each sub-agent should communicate and what prior subtask context it should see.
On benchmarks, Fugu Ultra scored 95.5% on GPQA Diamond (vs Claude Opus 4.8's 92.0%, GPT-5.5's 93.6%, Gemini 3.1 Pro's 94.3%), 73.7% on SWE-Bench Pro (vs Opus 4.8's 69.2%, GPT-5.5's 58.6%, Gemini 3.1 Pro's 54.2%), 93.2% on LiveCodeBench (vs Gemini 3.1 Pro's 88.5%), and 50.0% on Humanity's Last Exam (vs Opus 4.8's 49.8%). All scores are vendor-reported in Sakana's June 2026 technical report and have not been independently reproduced by third-party labs. The orchestration nature of Fugu means these scores reflect the capability of the whole coordinated system, not a set of single-model weights.
Latency is the trade-off. On simple prompts, Fugu adds only a small orchestration overhead and behaves like a single fast model. On hard prompts, multiple models are working in parallel or sequence before any output reaches the user, and latency can range from 11 seconds to over four minutes depending on problem complexity. The standard Fugu tier is the right choice for interactive work (chat, code review, quick searches). Fugu Ultra is for batch-mode research, complex engineering, and multi-step analysis where answer quality matters more than response time.
Pricing follows a usage-based structure for Fugu Ultra, a pass-through rate for the base Fugu tier, and optional monthly subscriptions bundling both; exact rates are in the cost FAQ answer below.
The API is OpenAI-compatible and served at https://api.sakana.ai/v1. Users authenticate with a Sakana API key. The model pool can be customized per API key: when creating or editing a key in the console at console.sakana.ai, users can specify which providers Fugu is allowed to route to. This is the primary mechanism for sovereignty control, allowing Japan-based enterprises to exclude specific US-origin models if needed.
Fugu was designed explicitly to operate without dependency on any single frontier model, specifically addressing the risk that Japan-based enterprises could lose access to Anthropic, OpenAI, or Google models under evolving US export controls. The underlying model pool is swappable without any API change, so if one provider restricts access, Sakana can substitute another. This sovereign positioning is unusual among orchestration systems and has been highlighted by Sakana as a core design goal rather than a side effect.
Sakana Fugu is best suited for teams that need GPQA-class reasoning or coding quality but cannot tolerate single-model API risk, or for tasks that genuinely benefit from multi-model verification: research synthesis, legal analysis, scientific reasoning, and complex engineering. It is not suitable for real-time voice assistants (latency variance is too high), for teams that require fully on-premise or air-gapped deployment (the model pool connects externally), or for workloads where per-token cost must be minimized (Fugu Ultra's output pricing is among the more expensive frontier options).
Pricing
Fugu Ultra standard tier: $5 input / $30 output / $0.50 cached per 1M tokens (up to 272K context). Extended context (272K to 1M): $10 / $45 / $1.00 per 1M tokens. Base Fugu tier charges the underlying model rate. Subscriptions: $20/month Standard, $100/month Pro, $200/month Max.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.150 | $0.030 | $0.180 |
| Support reply | $0.010 | $0.0090 | $0.019 |
| One coding agent run | $1.00 | $0.600 | $1.60 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Key Features
- Single-endpoint multi-model orchestration: One OpenAI-compatible API call internally routes to the optimal combination of frontier LLMs via the TRINITY and Conductor frameworks, returning one synthesized answer.
- Terminal-Bench 82.1% (vendor-reported): Fugu Ultra's highest published benchmark score is on Terminal-Bench, a test of autonomous terminal and agentic task completion, per Sakana's June 2026 technical report.
- 1,000,000-token extended context window: Fugu Ultra supports up to 1,000,000 tokens of context, priced across two context tiers detailed in the pricing FAQ below.
- Swappable sovereign model pool: Per-API-key provider restrictions, set in the Sakana console, let Japan-based teams exclude specific model providers without changing any application code.
- No-stacking flat pricing on multi-model calls: When multiple models coordinate on one request, Sakana bills only the single most expensive model in the chain rather than summing every model used.
Pros
- Vendor-reported benchmark leadership across GPQA Diamond and SWE-Bench Pro tops every single-model frontier API HokAI has tracked as of June 2026, pending independent verification.
- Sovereign routing removes the single-provider dependency risk that worries Japan-based enterprises facing evolving US export rules, without requiring any code change.
- Drop-in OpenAI SDK compatibility keeps migration cost near zero for teams already building on OpenAI's client libraries.
Cons
- Latency variance is severe on hard tasks: response times swing from seconds to minutes as the model pool coordinates in sequence or parallel.
- No third-party lab has confirmed any Fugu benchmark number yet, so procurement teams should treat every score as provisional.
- No deployment on AWS Bedrock, Google Vertex, or Azure; access is API-only through Sakana's own endpoint.
Benchmarks
- GPQA Diamond: 95.5% vendor-reported · 22 Jun 2026 — PhD-level science questions that are hard to search for, % correct.
- SWE-bench Pro: 73.7% vendor-reported · 22 Jun 2026 — Harder, longer real-repository coding tasks, % solved.
- Livecode Bench: 93.2 vendor-reported · 22 Jun 2026
- Terminal Bench: 82.1 vendor-reported · 22 Jun 2026
- Humanity's Last Exam: 50% vendor-reported · 22 Jun 2026 — Expert-written questions across many fields, % correct.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
How much do you pay for Sakana Fugu?
Fugu Ultra costs $5 per million input tokens and $30 per million output tokens for contexts up to 272,000 tokens, rising to $10 input and $45 output per million tokens for longer contexts, with cached input at $0.50 to $1.00 per million. The base Fugu tier has no flat rate: it bills at the price of whichever underlying model handles the request, and Sakana charges only for the single most expensive model when several are active. Monthly subscriptions bundling both tiers start at $20 (Standard), scaling to $100 (Pro) and $200 (Max).
Can you use Sakana Fugu without paying?
No. Sakana Fugu has no free tier, and every request, on the base Fugu tier or Fugu Ultra, requires a funded API key or an active subscription; there is no trial credit or sandbox key documented as of June 2026. The lowest-cost path in is the monthly Standard subscription described above, not pay-per-token billing.
What are the best alternatives to Sakana Fugu?
Claude Opus 4.8 and OpenAI's latest flagship model are the closest single-model competitors, both with lower latency variance than Fugu's multi-agent design. Gemini 3.1 Pro is a similar large-context option if a Google-hosted deployment through Vertex AI matters more than sovereign model-pool control. Teams that need open weights or self-hosting, which Fugu does not offer, are better served by Llama 4 or Mistral AI's open-source models.
Sakana Fugu or Claude Opus 4.8: which should you pick?
Fugu Ultra edges out Claude Opus 4.8 on Sakana's own reasoning and coding benchmarks, but every Fugu score is vendor-reported and has not been independently reproduced. On Humanity's Last Exam the gap is negligible: 50.0% for Fugu Ultra against 49.8% for Opus 4.8. The bigger practical difference is response time: Fugu Ultra's multi-agent coordination can take anywhere from seconds to minutes depending on task complexity, while a single-model API like Opus 4.8 responds in roughly constant time.
What does it take to start using Sakana Fugu?
Create an account at console.sakana.ai and generate an API key; the Fugu API at https://api.sakana.ai/v1 is OpenAI-compatible, so existing OpenAI SDK code works after swapping the base URL and key. Start with the base Fugu tier for everyday requests, and switch the model parameter to Fugu Ultra only for tasks that need the extra reasoning depth and can tolerate longer response times. Enterprises wanting sovereign routing should configure provider restrictions under the key's Providers settings in the console before sending production traffic.
Top Alternatives
- Claude Opus 4.8: Pick Sakana Fugu if you need GPQA-level reasoning without single-provider lock-in; pick Claude Opus 4.8 for lower latency and independently verified benchmarks.
- OpenAI: Pick Sakana Fugu if you need a swappable multi-model backend and Japan sovereign routing; pick OpenAI GPT-5.5 for faster latency and broader third-party benchmark coverage.