Also, from the technical report, looks like they're training on the output of Claude Code, etc. I'm guessing this doesn't violate TOS because they're technically not a directly competing model. This brings me to what I see as the main risk with this service, which is that it seems like an easy thing for a frontier lab to make obsolete, either by models beginning to converge in terms of strengths or by improving their own harnesses to include more of this meta-reasoning.
Does multiple vendors run this "single API" or how is this not replacing a single-vendor dependency for another single-vendor dependency?
it's interesting that they're offering in the form of fixed cost subscription plans too. My impression was that the first party providers can do this because they api inference margins to the tune of 80ish percent. Anyone else orchestrating on top of these models have to pass through these costs or eat it themselves.
If cost becomes an even bigger problem being able to choose "best performance possible" or "strong but cost effective" will be useful.
The reasoning chains could have been used, and the resulting combined model could easily and effectively have been distilled.
While you're at it, feel free to send me $200 as well, I'll generate a crypto address ending with "AI".
This gets you that in a nice neat package, without the underlying tinkering mechanics.
If (big iff) the usage mechanics work out, then this is actually a really good anti-big-model strategy.
They'll be incentivized for your success, not token-maximizing for their investors.
The team is super smart too. What's not to like?
Wishing them the best on launch.
Not yet available in the EU/EEA while we work toward compliance with GDPR and EU-specific regulations.https://japannews.yomiuri.co.jp/politics/defense-security/20...
(don't send anything, sharing only because of the base58 fun fact I didn't know)
I've been shipping production on archive.tw with Fugu Ultra in /advisor on oh-my-pi.
Advisor doesnât slow the loop if the driver stays fast. Worth it if your harness can split advisor from worker.
There's also the concept of "smart routing" requests based on some heuristics / embeddings. You'd get "simple" tasks handled by smaller (cheaper) models and use a bigger model to curate / sort / merge the results.
There's a lot of things to try here. I wouldn't personally pay for this service, but I don't think it's "a joke"...
Is there any official source that could confirms if Fable (or Mythos) is parallelized test-time compute (like GPT 5.5 Pro) or sparse Mixture-of-Experts (MoE) transformer combined with a multi-agent, inference-time compute scaling architecture (Gemini 3.1 Deep Think)?
This is ask a special orchestrator they built, which is in front of a bunch of models, which model would suit the request best.
Regular Fugu seems to be just "pick the best model and route the request there"
Fugu Ultra can generate like a little mini workflow/plan instead to achieve a result
1. Ask GPT to derive the math. 2. Ask Opus to check for implementation/security issues. 3. Ask Gemini to synthesize or resolve disagreement. 4. Return final answer.
I could be wrong but seems to be that at a glance, so I think it's more dynamic than OpenRouter Fusion.
> So basically... openrouter
:skull:
i now really wonder how many people of the public understood my thesis defense lol
We open sourced it all
and will be releasing a similar orchestrator next week on TrustedRouter
Looks like Fusion calls a bunch of models and then uses an LLM to synthesize the results, and pass to another model for final output.
Fugu looks like it's doing something different? Using an LLM earlier on in the flow as an orchestrator to decide which other LLMs to call. More coordinator than simply synthesizing results, and more "agentic".
It's interesting because it's all exposed behind a single OpenAI compatible endpoint (Responses API?) and so then presumably someone could use this for one of their single agents. Now you have agent-of-agents, nested in some sense. The token usage increases accordingly!
Basically, if you combine a bunch of near-frontier models (like GPT 5.5, etc) you can get performance that sometimes surpasses top line models like Claude's Fable.
Sakana seems to have a separate approach using a domain specific model to perform the model routing step.
These prices are just going to get raced to $0.
But their paid plans I'm not sure yet - planning to subscribe and can let you know.
Almost no chance it will be as generous as OpenAI though. They just don't have the money :-)
It's similar to how AirPods normalised all of us having $300+ headphones. All of us would have scoffed at the idea a decade ago.
Guess what, the big players are hoarding all the RAM and GPUs so that other people can't afford decent hardware. It's working out beautifully for them!
It's $200/month. You have to take into account energy costs and all the rest of a system, but if you break even within 1-2 years ($2400-$4800) it'd be a pretty good deal. And $4000 buys you a pretty decent system.
EDIT: Found something here https://dev.classmethod.jp/en/articles/sakana-fugu-ga-first-...
All put together, paying ~$60 to get a hit-or-miss report seems a bit excessive, but obviously as the models they use under the hood get better it becomes more and more worth it, assuming they also improve their grounding/search capabilities.
I'm a big fan of Sakana though, and have followed David Ha / @hardmaru since the world models papers (with the racing car game and the Doom clone), which were incredible at the time.
After a few months of spending money on the best frontier models, now I am spending time using DeepSeek v4 flash as my workhorse, and flipping to more capable (but still very inexpensive) open models on an as-needed basis. We all make our own tool selection decisions, but for me, I feel happier and enjoy working more following the very fast response and ultra low cost path.
David Ha, CEO and co-founder, was one of the youngest managing director at Goldman Sachs before doing ML at Google. His ML publications were considered top-notch almost a decade ago. I had high hopes for him when he raised money and founded Sakana.
I do agree with some comments here that perhaps this particular product is not well thought out. I also agree with the criticism that David calls Sakana a frontier AI lab while making money just selling AI B2B applications to Japanese businesses. I also agree with the assessment that Sakana has abrasive and antagonistic, sometimes openly hostile, recruiting tactics. I also agree that his then-impressive publications may have lost their luster in the age of LLMs.
However, the man is clearly driven; and he and his team may have more to offer in future. I admire the man for not taking the conventional AI-research career path.
They are talking about a much larger group of people.
Personally I prefer understanding the dimensions and the interplay and controlling it though can see why openrouter and others are now offering this a solved solution.
Just be careful when you start outsourcing too much of your intelligence needs to a blackbox.
The same model that has been post-trained to operate for hours as a Linux admin will be incapable of writing a heartfelt email, but with something like Fugu, you'd get both the Linux admin for driving the browser harness and the smaller writing specialist model for drafting the email itself.
For others looking around: LCF is a meme model, it's not real. It's a joke.
https://openai.com/index/our-agreement-with-the-department-o...
https://www.databricks.com/blog/introducing-omnigent-meta-ha...
More broadly, Sakana is pursing a refreshingly distinct research path, with their focus on evolutionary methods, biological intelligence (e.g. continuous thought machines) and open publication.
Like every company based in China they are under the control of the Chinese state, which is an armed entity known to use violence.
But your aunt Josie didn't have one. Now Apple is selling 80 million units / year and the ~$300 price tag has become normal. Before that, most people had headphones that were 10 times cheaper.
Probably taking hate from both sides - OpenAI / Claude fans who are undercutting its moat. Chinese open-model fans that want it to be cheaper.
But it's a genuine accomplishment to hit those benchmarks and offer a reasonable plan?
Bizarre reaction TBH.
At least, for the initial data gathering phase. You'd probably want a sequence of progressively larger models to filter it.
Have you guys tested it on anything other than research?
I just averaged it out.
[0] https://dev.classmethod.jp/en/articles/sakana-fugu-ga-first-...
One Model to Command Them All ãã«ããšãŒãžã§ã³ããææ®ãããäžã€ã®ã¢ãã«
Frontier-level performance without single-vendor dependency. Fugu dynamically orchestrates the world's best models to tackle complex, multi-step tasks. Plug collective intelligence directly into your workflows today with a single API. Sakana Fugu ã¯ãäžçã®ãããã¢ãã«çŸ€ãåçã«ãªãŒã±ã¹ãã¬ãŒã·ã§ã³ããè€æ°ã¹ãããã«åã¶è€éãªã¿ã¹ã¯ãèªåçã«è§£æ±ºããŸããé«ãããã©ãŒãã³ã¹ãå®çŸããAPIããããªãã®ã¯ãŒã¯ãããŒã«çµã¿èŸŒã¿ãŸãããã
Not yet available in the EU/EEA while we work toward compliance with GDPR and EU-specific regulations. GDPRçã®EU/EEAåºæèŠå¶ãžã®å¯Ÿå¿ãé²ããŠãããçŸåšã¯EUã»EEAåå ã§ã¯ãå©çšããã ããŸããã
What is Sakana Fugu ?
Sakana Fugu achieves superior performance by dynamically coordinating and orchestrating a diverse pool of powerful models. Instead of using domain knowledge to prescribe team organization, roles, or workflows, Fugu learns to dynamically assemble agents from a pool and coordinate them through non-obvious but highly efficient collaboration patterns. Sakana Fugu ã¯ã匷åã§å€æ§ãªã¢ãã«çŸ€ãåçã«çµã¿åãããå調ãããããšã§é«ãããã©ãŒãã³ã¹ãå®çŸããŸãã人éãæãä»ããªããããªã¢ãã«ã®ç·šæã圹å²åæ ãåŠçã®é²ãæ¹ãªã©ãå¹çããåŠç¿ããªããææãçºæ®ããŸãã

01
Access a coordinated pool of specialized models through one API. Fugu handles model selection and switching for each task, reducing API complexity while improving cost-performance. å°éç¹ååã®ã¢ãã«çŸ€ããäžã€ã®APIããå©çšããããšãã§ããŸããã¿ã¹ã¯ããšã®ã¢ãã«ã®éžæãšåãæ¿ã㯠Sakana Fugu ãæ ããããAPIãŸããã®ç ©éããæãã€ã€ãã³ã¹ãããã©ãŒãã³ã¹ãé«ããããŸãã
02
Built for coding, reasoning, and other quality-critical workflows, Fugu coordinates expert agents to tackle complex tasks with stronger, more reliable results. Sakana Fugu ã¯ãã³ãŒãã£ã³ã°ãæšè«ïŒãªãŒãºãã³ã°ïŒãªã©ãé«ãå質ãåãããã¯ãŒã¯ãããŒã®ããã«èšèšãããŠããŸããå°éãšãŒãžã§ã³ãã飿ºãããããšã§ãè€éãªã¿ã¹ã¯ã«ããã確ãã§ä¿¡é Œã§ããçããå°ããŸãã
03
Control which agents can participate in Fuguâs model pool. Opt out of specific providers or models to meet data, privacy, compliance, or organizational requirements. Sakana Fugu ã®ã¢ãã«ããŒã«ã«å ãããšãŒãžã§ã³ããéžã¶ããšãã§ããŸããããŒã¿ããã©ã€ãã·ãŒãã³ã³ãã©ã€ã¢ã³ã¹ããŸãã¯çµç¹ã®èŠä»¶ãæºããããã«ãç¹å®ã®ãããã€ããŒãã¢ãã«ãé€å€ããããšãå¯èœã§ãã
Tech Behind
Sakana Fugu is grounded in two ICLR 2026 papers on learned model orchestration: TRINITY and the Conductor. Together, they show how systems can learn to assemble, route, and coordinate expert agents for each task instead of relying on hand-designed workflows. For a deeper look at the ideas behind the system, explore our technical report . Sakana Fugu ã¯ãã¢ãã«ã®ãªãŒã±ã¹ãã¬ãŒã·ã§ã³ãåŠç¿ã§å®çŸãã2æ¬ã®ICLR 2026è«æãTRINITYããšãConductorããåºç€ãšããŠããŸãããããã®ç ç©¶ã¯ã人æã§èšèšããã¯ãŒã¯ãããŒã«é Œãã®ã§ã¯ãªããã¿ã¹ã¯ããšã«å°éãšãŒãžã§ã³ããã©ãç·šæããæ¯ãåãã飿ºãããããã·ã¹ãã èªèº«ãåŠç¿ã§ããããšã瀺ããŠããŸããä»çµã¿ã®è©³çްã¯ã ãã¯ãã«ã«ã¬ããŒã ãã芧ãã ããã

Trinity uses a lightweight evolved coordinator to orchestrate multiple LLMs over several turns, assigning Thinker, Worker, or Verifier roles to adaptively delegate work across coding, math, reasoning, and knowledge tasks. TRINITY ã¯ã軜éãªé²ååã³ãŒãã£ããŒã¿ãŒãè€æ°ã®LLMãè€æ°ã¿ãŒã³ã«ããã£ãŠçµ±æ¬ããä»çµã¿ãåã¢ãã«ã«ãThinkerïŒæè圹ïŒããWorkerïŒå®è¡åœ¹ïŒããVerifierïŒæ€èšŒåœ¹ïŒãã®åœ¹å²ãå²ãåœãŠãã³ãŒãã£ã³ã°ã»æ°åŠã»æšè«ã»ç¥èãšãã£ãå¹ åºãã¿ã¹ã¯ã«å¿ããŠãäœæ¥ãé©å¿çã«æ¯ãåããã

The Conductor is trained with reinforcement learning to discover natural-language coordination strategies, designing agent communication patterns and focused prompts that help diverse LLM pools outperform individual workers on challenging reasoning benchmarks. Conductor ã¯åŒ·ååŠç¿ã«ãã£ãŠèšç·Žãããèªç¶èšèªããŒã¹ã®å調æŠç¥ãèªãèŠã€ãåºãããšãŒãžã§ã³ãéã®ããåãã®åããèŠç¹ãçµã£ãããã³ãããèšèšããããšã§ã倿§ãªLLMã®éãŸãããé£åºŠã®é«ãæšè«ãã³ãããŒã¯ã§åäœã®ã¢ãã«ãäžåãåãçºæ®ã
How to Use
Sakana Fugu comes in two models â Fugu and Fugu Ultra â both available through one OpenAI-compatible API. Pick the model that fits your workload, or switch between them without changing your integration. Sakana Fugu ã«ã¯ Fugu ãš Fugu Ultra ã® 2 ã€ã®ã¢ãã«ããããã©ã¡ãã OpenAI äºæ API ããå©çšã§ããŸããã¯ãŒã¯ããŒãã«åãã¢ãã«ãéžãã§ãã飿ºãå€ããã«äž¡è ãåãæ¿ããŠãããŸããŸããã
Fugu Balanced performance and latency æ§èœãšã¬ã€ãã³ã·ã®ãã©ã³ã¹
Fugu balances strong performance with low latency, making it the ideal default for everyday work. Drop it into tools like Codex for coding and code review, or power responsive chatbot services â all behind a single endpoint. You can also opt specific agents out of its pool to meet data, privacy, and compliance constraints. Sakana Fugu ã¯é«ãæ§èœãšäœã¬ã€ãã³ã·ãäž¡ç«ããæ¥ã ã®äœæ¥ã«æé©ãªæšæºã¢ãã«ã§ããCodex ã®ãããªããŒã«ã«çµã¿èŸŒãã§ã³ãŒãã£ã³ã°ãã³ãŒãã¬ãã¥ãŒã«äœ¿ã£ãããå¿çæ§ã®é«ããã£ããããããåããããââãã¹ãŠãã²ãšã€ã®ãšã³ããã€ã³ãã§å®çŸããŸããããŒã¿ã»ãã©ã€ãã·ãŒã»ã³ã³ãã©ã€ã¢ã³ã¹ã®å¶çŽã«åãããŠãããŒã«ããç¹å®ã®ãšãŒãžã§ã³ããé€å€ããããšãã§ããŸãã
Fugu Ultra Optimized for performance æ§èœã«æé©å
Fugu Ultra coordinates a deeper pool of expert agents to maximize answer quality on hard, high-stakes problems. Early users rely on it for Kaggle competitions, paper reproduction, cybersecurity analysis, and literature and patent investigations. Fugu Ultra ã¯ãããåºãå°éãšãŒãžã§ã³ãã®ããŒã«ã飿ºãããé£æåºŠãé«ãéèŠãªåé¡ã§åçå質ãæå€§åããŸããå è¡ãŠãŒã¶ãŒã¯ãKaggle ã³ã³ããã£ã·ã§ã³ãè«æã®åçŸããµã€ããŒã»ãã¥ãªãã£åæãæç®ã»ç¹èš±èª¿æ»ãªã©ã«æŽ»çšããŠããŸãã
Quantitative Results
Our Fugu models surpass publicly accessible frontier models and are shoulder-to-shoulder with Fable 5 and Mythos Preview in various rigorous engineering, scientific, and reasoning benchmarks while delivering frontier capability without the risk of export controls. äºã€ã®Fuguã¢ãã«ã¯ãäžè¬ã«å©çšã§ããããã³ãã£ã¢ã¢ãã«ãäžåãããšã³ãžãã¢ãªã³ã°ã»ç§åŠã»æšè«ã®ããŸããŸãªé£é¢ãã³ãããŒã¯ã§ããFable 5ãMythos Previewãšè©ã䞊ã¹ãŸãããããã茞åºèŠå¶ã®ãªã¹ã¯ãè² ãããšãªããããã³ãã£ã¢ã¬ãã«ã®å®åãçºæ®ããŸãã

Performance comparison of Fugu models and baseline frontier models across a suite of coding, reasoning, scientific, and agentic benchmarks. For Fable 5 and Mythos Preview, we report the max of the two if both scores are available on the same benchmark. Neither of them is in Fuguâs agent pool as they are not publicly accessible. ã³ãŒãã£ã³ã°ããªãŒãºãã³ã°ãç§åŠããšãŒãžã§ã³ãèœåã«é¢ãããã³ãããŒã¯çŸ€ã«ããããFuguã¢ãã«ãšããŒã¹ã©ã€ã³ã®ããã³ãã£ã¢ã¢ãã«ã®æ§èœæ¯èŒãFable 5ãšMythos Previewã«ã€ããŠã¯ãåäžãã³ãããŒã¯ã§äž¡æ¹ã®ã¹ã³ã¢ãå ¥æã§ããå Žåããã®é«ãæ¹ãæ¡çšããªããäž¡ã¢ãã«ã¯äžè¬æäŸãããŠããªããããFuguã®ãšãŒãžã§ã³ãããŒã«ã«ã¯å«ãŸããŠããªãã
Highest scores are shown in boldface; second-highest scores are underlined. æé«ã¹ã³ã¢ã¯å€ªåã2 çªç®ã«é«ãã¹ã³ã¢ã¯äžç·ã§ç€ºããŠããŸãã
| Benchmark | Fugu | Fugu Ultra | Opus 4.8 â | Gemini 3.1 Pro â | GPT 5.5 â |
|---|---|---|---|---|---|
| SWE Bench Pro * | 59.0 | 73.7 | 69.2 | 54.2 | 58.6 |
| TerminalBench 2.1 | 80.2 | 82.1 | 74.6 | 70.3 | 78.2 |
| LiveCodeBench | 92.9 | 93.2 | 87.8 | 88.5 | 85.3 |
| LiveCodeBench Pro | 87.8 | 90.8 | 84.8 | 82.9 | 88.4 |
| Humanityâs Last Exam | 47.2 | 50.0 | 49.8 | 44.4 | 41.4 |
| CharXiv Reasoning | 85.1 | 86.6 | 84.2 | 83.3 | 84.1 |
| GPQA-D | 95.5 | 95.5 | 92.0 | 94.3 | 93.6 |
| SciCode | 60.1 | 58.7 | 53.5 | 58.9 | 56.1 |
| ϳ Banking | 21.7 | 20.6 | 20.6 | 8.4 | 20.6 |
| Long Context Reasoning | 74.7 | 73.3 | 67.7 | 72.7 | 74.3 |
| MRCRv2 | 86.6 | 93.6 | 87.9 | 84.9 | 94.8 |
* We use the mini-swe-agent as the scaffolding for this task. * mini-swe-agent ãã¹ãã£ãã©ãŒã«ããšããŠäœ¿çšã
â We use model provider-reported scores for the baselines. â ã¢ãã«æäŸå ãå ¬è¡šããã¹ã³ã¢ã
Qualitative Results
These examples compare Sakana Fugu models with three frontier baselines â Gemini 3.1 Pro (high) , Opus 4.8 (max) , and GPT 5.5 (xhigh) . To keep the focus on behavior rather than brand-by-brand attribution, the baselines are anonymized as Model A , Model B , and Model C in each description. The mapping is intentionally not fixed across examples. 以äžã®äŸã§ã¯ã Sakana Fugu ãã Gemini 3.1 ProïŒhighïŒ ã Opus 4.8 ïŒmaxïŒ ã GPT 5.5ïŒxhighïŒ ã®3ã€ã®ããã³ãã£ã¢ã¢ãã«ãšæ¯èŒããŠããŸããåå¥ã¢ãã«ã§ã¯ãªãæåã®éãã«æ³šç®ã§ãããããããŒã¹ã©ã€ã³ã Model A ã Model B ã Model C ãšããŠå¿ååããŠããŸãã ãªããã©ã®ã¢ãã«ãAãCãã¯äŸããšã«å€ããŠããŸãã
This experiment shows an AI agent autonomously improving a small GPT's training recipe. Using AutoResearch (Karpathy et al.) â which iteratively edits training code, runs experiments, and keeps only changes that lower validation bits-per-byte (BPB) â the agent ran 123 experiments over ~14 hours on a single H100 GPU. Each line traces a system's best BPB as experiments accumulate: Fugu-Ultra is in bold red (solid = mean over three seeds, dashed = best single run), with three frontier-model baselines (Model A, B, and C) faded behind it, and the callouts mark each new improvement the agent found on its own â spanning batch size, model depth, learning rates, and optimizer settings. Fugu-Ultra finishes with the best mean BPB (0.9774 ± 0.0019), ahead of Model C (0.9781), Model B (0.9793), and Model A (0.9822), and its best single run reaches 0.9748, leading every baseline. This suggests that orchestrating multiple strong models can outperform any individual frontier model on agentic ML research. äŸ1 â AutoResearch / LLMåŠç¿
AIãšãŒãžã§ã³ãã«å°èŠæš¡ãªGPTã®åŠç¿ã¬ã·ããèªåŸçã«æ¹åãããå®éšãåŠç¿ã³ãŒããå埩çã«æžãæããå®éšãå®è¡ããæ€èšŒçš bits-per-byteïŒBPBïŒãäžãã倿Žã ããæ®ããŠãããšãŒãžã§ã³ãåãã¬ãŒã ã¯ãŒã¯ AutoResearchïŒKarpathy et al.ïŒãçšãããšãŒãžã§ã³ãã¯åäžã®H100 GPUäžã§ããã14æéã«ããã123åã®å®éšã宿œãããåç·ã¯ãå®éšãç©ã¿éãªãã«ã€ããŠåã·ã¹ãã ãéæããæè¯ã®BPBã®æšç§»ã衚ããŠãããFugu-Ultra ã¯å€ªãèµ€ã®ç·ïŒå®ç·ïŒ3ã·ãŒãã®å¹³åãç Žç·ïŒæè¯ã®åäžå®è¡ïŒã§ç€ºãããã®èåŸã«3ã€ã®ããã³ãã£ã¢ã¢ãã«ã®ããŒã¹ã©ã€ã³ïŒModel Aã»Bã»CïŒãæ·¡è²ã§éããŠãããå¹ãåºãã¯ããšãŒãžã§ã³ããèªãèŠã€ããæ¹åç¹ããããã瀺ããŠãããããããµã€ãºãã¢ãã«ã®æ·±ããåŠç¿çããªããã£ãã€ã¶ã®èšå®ãªã©å€å²ã«ããããFugu-Ultra ã¯æçµçã«æè¯ã®å¹³åBPBïŒ0.9774 ± 0.0019ïŒãéæããModel CïŒ0.9781ïŒãModel BïŒ0.9793ïŒãModel AïŒ0.9822ïŒãäžåã£ããæè¯ã®åäžå®è¡ã§ã¯ 0.9748 ã«å°éãããã¹ãŠã®ããŒã¹ã©ã€ã³ãäžåã£ãŠããããããã®çµæã¯ãè€æ°ã®åŒ·åãªã¢ãã«ããªãŒã±ã¹ãã¬ãŒã·ã§ã³ããããšã§ããšãŒãžã§ã³ãåã®MLç ç©¶ã«ãããŠåäœã®ããã³ãã£ã¢ã¢ãã«ãäžåãåŸãããšã瀺åããŠããã
This case study tests whether the reading order of classical Japanese kana letters (仮忶æ¯) can be recovered â letters whose scattered chirashigaki ("scattered-writing") layout makes that genuinely hard even for trained readers of classical Japanese. Each model is given the character bounding boxes together with a rough set of reading-order rules, and writes code that outputs the order the characters should be read in; here it runs on a letter written in 1610 by HÅshun'in (è³æ¥é¢, 1547â1617), scored by NED (a score based on normalized edit distance from an expert's ground-truth order, where 1.0 is a perfect match). Several frontier models were put through the identical pipeline, but none came close to Fugu-Ultra on this letter: Model A reached only NED 0.24 and Model B scored no better, both far below Fugu-Ultra's 0.80, while Model C produced no predictor at all. The clip shows the two extremes â each panel draws its predicted path in red over the expert's ground truth in green: Fugu-Ultra (top) traces the letter almost exactly, while Model A (bottom) jumps all over the page. (Letter held by the Keio Institute of Oriental Classics.) äŸ2 â 仮忶æ¯ã®èªã¿é æšå®
æ¬ã±ãŒã¹ã¹ã¿ãã£ã¯ã仮忶æ¯ïŒå€å
žæ¥æ¬èªã®ããªæžç¶ïŒãšããæŽå²çè³æã«ãããèªã¿é ã®æšå®åé¡ã察象ãšããã仮忶æ¯ã¯ãæåãçŽé¢ã«æ£ãããŠèšããæ£ããæžãããšãã圢åŒã§æžãããŠããããã倿æžãèªã¿æ
£ãã人ã§ãæåã®èªã¿é ãæ£ããå€å®ããããšã¯é£ãããããã§åã¢ãã«ã«å¯ŸããŠãæåãå²ãåè§åœ¢ïŒããŠã³ãã£ã³ã°ããã¯ã¹ïŒãšèªã¿é ã®å€§ãŸããªã«ãŒã«ãäžããæåã®èªã¿é ãæšå®ããã³ãŒããåºåããããå®éšã®å¯Ÿè±¡ã«ã¯1610幎ã«è³æ¥é¢ïŒã»ããã
ãããã1547â1617ïŒãèšããæžç¶ãéžã³ãNEDïŒå°éå®¶ã«ããæ£ããèªã¿é ãšã®æ£èŠåç·šéè·é¢ã«ããšã¥ãã¹ã³ã¢ã1.0ãå®å
šäžèŽïŒã§è©äŸ¡ãããè€æ°ã®ããã³ãã£ã¢ã¢ãã«ïŒA-CïŒãåäžã®ãã€ãã©ã€ã³ã«éãããšãããFugu-Ultraã®çµæã¯ä»ã®ã¢ãã«ã倧ããåŒãé¢ãããModel Aã¯NED 0.24ãModel Bããããšå€§å·®ãªããããããFugu-Ultraã®0.80ã«ã¯é ãåã°ãªããããã«Model Cã¯ãŸãšããªã³ãŒããäžåãåºåã§ããªãã£ããã¢ãã«ã«ããèªã¿é ã®éããå¯èŠåããããã«ãå°éå®¶ã«ããæ£è§£ã®èªã¿é ïŒç·ïŒã®äžã«ãæšå®ããçµè·¯ïŒèµ€ïŒãæããŠæ ååãããFugu-UltraïŒäžïŒãèªã¿é ãã»ãŒæ£ç¢ºã«ãªããäžæ¹ãModel AïŒäžïŒã¯çŽé¢å
šäœããã¡ãã¡é£ã³åããäž¡è
ã¯å€§ããç°ãªãçµæã瀺ããŠããã å³ïŒè³æ¥é¢æ¶æ¯ïŒæ
¶æçŸ©å¡Ÿå€§åŠæ¯éæåº«èµïŒ
In this benchmark, each of Fugu-Ultra and 3 frontier models is given a single prompt to write a Rubik's Cube solver from scratch in pure Python â no off-the-shelf solving libraries allowed â and the resulting program is run locally on a held-out set of 300 randomly scrambled cubes. Solution quality is measured by the number of moves a solution uses, where lower is better. Fugu-Ultra and the frontier Model A wrote solvers that ran and solved all 300 cubes, while Model B and Model C each shipped sophisticated-looking code that crashed on execution and returned no valid solution at all (0/300). The clip follows cube #17: from the same scramble, Fugu-Ultra's solver reaches the solved state in 19 moves while Model A needs 21 â and across all 300 cubes Fugu-Ultra averages 19.72 moves versus 19.76 for Model A, both right at the optimal frontier, with Fugu-Ultra never a move longer than Model A on any cube (7 wins, 293 ties, 0 losses). äŸ3 â ã«ãŒããã¯ãã¥ãŒãã»ãœã«ããŒ
æ¬ãã³ãããŒã¯ã§ã¯ãFugu-Ultraãš3ã€ã®ããã³ãã£ã¢ã¢ãã«ããããã«ãçŽç²ãªPythonã®ã¿ã§ã«ãŒããã¯ãã¥ãŒããœã«ããŒããŒãããå®è£
ããããåäžã®ããã³ãããäžãããæ¢åã®ãœã«ããŒã©ã€ãã©ãªã®äœ¿çšã¯çŠæ¢ãšããçæãããããã°ã©ã ãã©ã³ãã ã«ã¹ã¯ã©ã³ãã«ããã300åã®ãã¥ãŒããããªãããŒã«ãã¢ãŠãã»ããã«å¯ŸããŠããŒã«ã«ã§å®è¡ãããè§£æ³ã®è³ªã¯ææ°ã§è©äŸ¡ããå°ãªãã»ã©è¯ããšãããFugu-Ultraãšããã³ãã£ã¢ã® Model A ã¯300åãã¹ãŠã®ãã¥ãŒããè§£ããœã«ããŒãçæããããModel B ãš Model C ã¯äžèŠæŽç·Žãããã³ãŒããåºåãããã®ã®ãå®è¡æã«ã¯ã©ãã·ã¥ããæå¹ãªè§£ãäžã€ãè¿ããªãã£ãïŒ0/300ïŒãæ åã¯ãã¥ãŒã#17ã®æ§åã§ãããåäžã®ã¹ã¯ã©ã³ãã«ã«å¯ŸããFugu-Ultraã®ãœã«ããŒã¯19æã§å®æç¶æ
ã«å°éããã®ã«å¯ŸããModel A ã¯21æãèŠããã300åå
šäœã®å¹³åã§ã¯ãFugu-Ultraã19.72æãModel A ã19.76æãšãããããæé©è§£ã®æ°Žæºã«ãããFugu-Ultraã Model A ããææ°ãå€ãã£ãã±ãŒã¹ã¯äžåºŠããªãã£ãïŒ7åã»293åŒãåãã»0æïŒã
Task: Create a mechanical iris in CAD, like a camera aperture, where multiple blades move together to open and close the central hole. For each model, we show both the generated detailed CAD itself and a simplified view that makes the structure easier to see. In the CAD generated by Fugu Ultra, the blades rotate around outer pins and clearly open and close the aperture. In contrast, the CAD generated by the other models shows problems such as gaps appearing, weak linkages, or the aperture not closing fully. äŸ4 â CAD ã¡ã«ãã«ã«ã¢ã€ãªã¹
ã¿ã¹ã¯ïŒã«ã¡ã©ã®çµãïŒã¢ããŒãã£ïŒã®ãããªãè€æ°ã®çŸœæ ¹ãé£åããŠåãäžå€®ã®ç©Žãééããæ©æ¢°åŒã¢ã€ãªã¹ãCADã§äœæãããåã¢ãã«ã«ã€ããŠãçæããã詳现CADïŒDetailed CADïŒãã®ãã®ãšãæ§é ãèŠãããããããã®ç°¡æãã¥ãŒïŒSimplified viewïŒã®äž¡æ¹ã瀺ããFugu UltraãçæããCADã§ã¯ãçŸœæ ¹ãå€åŽã®ãã³ã軞ã«å転ããã¢ããŒãã£ãæç¢ºã«ééã§ããŠãããäžæ¹ãä»ã®ã¢ãã«ãçæããCADã§ã¯ãééãã§ããŠããŸãããªã³ã¯æ©æ§ã匱ããã¢ããŒãã£ãååã«éããããŠããªãããšãã£ãåé¡ãèŠãããã
Four blindfold chess games, back to back. Every model plays the same way â no board shown â holding the full game in memory. Fugu outplays four strong opponents: three leading frontier models and a 2100-Elo Stockfish engine, staying accurate where they drift and ending each game in checkmate. äŸ5 â ç®é ããã§ã¹
4å±ã®ç®é ããã§ã¹ãé£ç¶ããŠå¯Ÿå±ããŠããããã¹ãŠã®ã¢ãã«ã¯åãæ¡ä»¶ã§ãã¬ã€ããç€é¢ã¯äžå衚瀺ããããã²ãŒã å
šäœãèšæ¶ã®äžã«ä¿æããªããæãæãé²ãããFugu ã¯4ã€ã®åŒ·åãªçžæââ3ã€ã®äž»èŠãªããã³ãã£ã¢ã¢ãã«ãšã2100-Elo ã® Stockfish ãšã³ãžã³ââãæã¡è² ããããçžæãæãä¹±ããŠããå Žé¢ã§ãæ£ç¢ºããä¿ã¡ããããã®å¯Ÿå±ããã§ãã¯ã¡ã€ãã§çµããã
This benchmark uses a single anonymized equity over one historical 50-week window and is intended to compare sequential, no-look-ahead decision-making rather than to establish generalizable trading performance. Past performance does not guarantee future results, and results may not transfer to other assets, time periods, or live markets. Each model makes online trading decisions on anonymized STOCK_X, using only current and past weekly market data: opening, high, low, and closing prices, volume, returns, moving averages, volatility, drawdown, portfolio state, and prior feedback. Starting with $10,000, the agent chooses whether to buy, hold, or sell, and what fraction of cash or shares to trade. After each action, the next week's price is revealed and the portfolio is updated, so the model must adapt from feedback rather than seeing the future. Across five runs of the identical 50-week pipeline, Fugu-Ultra grew the portfolio to $11,943.22 ± $633.86, a +19.43% mean return, while the other frontier models reached their return less than +15%. äŸ6 â æ ªåŒãã¬ãŒãã£ã³ã°
å¿ååãããåäžéæã1ã€ã®éå»50é±éã®ãŠã£ã³ããŠã§çšãããã®æ ªåŒãã¬ãŒãã£ã³ã°ã®ãã³ãããŒã¯ãæ±çšçãªãã¬ãŒãã£ã³ã°æ§èœãç«èšŒããããã§ã¯ãªããå
èªã¿ã®ãªã鿬¡çãªæææ±ºå®ãæ¯èŒããããšãç®çãšããŠãããéå»ã®å®çžŸã¯å°æ¥ã®çµæãä¿èšŒãããã®ã§ã¯ãªããçµæãä»ã®è³ç£ã»æéã»å®éã®åžå Žã«åœãŠã¯ãŸããšã¯éããªããåã¢ãã«ã¯ãå¿ååããã STOCK_X ã«å¯ŸããŠãçŸåšããã³éå»ã®é±æ¬¡ããŒã±ããããŒã¿ââå§å€ãé«å€ãå®å€ãçµå€ãåºæ¥é«ããªã¿ãŒã³ãç§»åå¹³åããã©ãã£ãªãã£ããããŒããŠã³ãããŒããã©ãªãªã®ç¶æ
ãçŽåã®ãã£ãŒãããã¯ââã®ã¿ãçšããŠãªã³ã©ã€ã³ã§ãã¬ãŒãã£ã³ã°ã®æææ±ºå®ãè¡ãã1äžãã«ããã¹ã¿ãŒããããšãŒãžã§ã³ãã¯è²·ãã»ä¿æã»å£²ãã®ãããããšãçŸéãŸãã¯æ ªåŒã®ã©ã®å²åãååŒããããéžæãããåã¢ã¯ã·ã§ã³ã®åŸã«ç¿é±ã®äŸ¡æ Œãé瀺ãããããŒããã©ãªãªãæŽæ°ããããããã¢ãã«ã¯æªæ¥ãèŠãã®ã§ã¯ãªããã£ãŒãããã¯ããé©å¿ããªããã°ãªããªããåäžã®50é±éãã€ãã©ã€ã³ã5åå®è¡ããçµæãFugu-Ultra ã¯ããŒããã©ãªãªã 11,943.22 ± 633.86 ãã«ãŸã§æé·ãããå¹³åãªã¿ãŒã³ã¯ +19.43% ã«éãããäžæ¹ãä»ã®ããã³ãã£ã¢ã¢ãã«ã®ãªã¿ãŒã³ã¯ãããã +15% æªæºã«ãšã©ãŸã£ãã
Users' Voices
01
Software Engineer ãœãããŠã§ã¢ãšã³ãžãã¢
For code review, Fugu Ultra is significantly better than GPT-5.5. It gives comprehensive answers and finds the bugs others miss. Where other tools flag about three issues, Sakana Fugu surfaced more than twenty. It's become the model I run all my reviews through. ã³ãŒãã¬ãã¥ãŒã§ã¯ãFugu Ultra ã¯åçãç¶²çŸ çã§ãä»ã®ã¢ãã«ãèŠéããã°ãŸã§èŠã€ããŠãããŸãããä»ã®ããŒã«ã§ã¯3ä»¶ãããã®åé¡ããææãããªãã£ããã Sakana Fugu ã¯20件以äžãæŽãåºããŠãããŸããã
02
Researcher (industry) ç ç©¶è ïŒäŒæ¥ïŒ
I was mapping a patent landscape across ~20 papers and several patents, normally 3â4 days of work. With Fugu I had a full analysis in a few hours, including connections between papers I would never have spotted on my own. çŽ20æ¬ã®è«æãšè€æ°ã®ç¹èš±ã«ãŸãããç¹èš±ååïŒããã³ãã©ã³ãã¹ã±ãŒãïŒãäœæããŸãããæ®æ®µãªã3ã4æ¥ãããäœæ¥ãã Sakana Fugu ã䜿ããšæ°æéã§å®å šãªåæãã§ãããã®ãªãã«ã¯ãèªåã§ã¯æ±ºããŠæ°ã¥ããªãã£ãã ããè«æå士ã®ã€ãªãããèŠã€ããããšãã§ããŸããã
03
Executive (enterprise platform) ãã©ãããã©ãŒã äŒæ¥ã»åœ¹å¡
Raw output quality is on par with top frontier models, but Fugu showed unusually strong persona stability across long sessions, holding its identity where other models drift. For agent products, that may matter more than raw benchmark scores. çŽ ã®åºåå質ã¯ãããã¯ã©ã¹ã®ããã³ãã£ã¢ã¢ãã«ãšåçãå ã㊠Sakana Fugu ã¯ãé·æéã®ã»ãã·ã§ã³ã§ããã«ãœããå®å®ããŠãããä»ã®ã¢ãã«ãªã厩ããŠããŸãå Žé¢ã§ããã£ã©ã¯ã¿ãŒãä¿ã¡ç¶ããŸããããšãŒãžã§ã³ãã«ãšã£ãŠã¯ãããã¯åçŽãªãã³ãããŒã¯ã¹ã³ã¢ä»¥äžã«éèŠãªããšã§ãã
04
Researcher ç ç©¶è
From one simple request, Sakana Fugu worked autonomously for nearly four hours â reading the paper, implementing, training, evaluating, and analyzing the gaps. äžã€ã®ã·ã³ãã«ãªæç€ºããã Sakana Fugu ã¯ããã4æéç¶ããŠèªåŸçã«äœæ¥ããŸãããè«æãèªã¿èŸŒã¿ãå®è£ ã»åŠç¿ã»è©äŸ¡ãŸã§è¡ããè¶³ããªãç¹ãåæããŠãããŸããã
05
Security Engineer ã»ãã¥ãªãã£ãšã³ãžãã¢
Given one scoped instruction, Sakana Fugu drove a full security assessment end-to-end â recon, XSS/SQLi checks, auth review, and a clean report with evidence and retest steps â staying inside scope and avoiding destructive actions. ç¯å²ãçµã£ãæç€ºãäžã€æž¡ããã ãã§ã Sakana Fugu ã¯æ å ±åéãã XSS/SQLi ã®æ€æ»ãèªèšŒãŸããã®ã¬ãã¥ãŒãããã«èšŒæ ãšåãã¹ãæé ãåããæŽç¶ãšããã¬ããŒãäœæãŸã§ãã»ãã¥ãªãã£è©äŸ¡ãäžæ°é貫ã§ããªããŸããããããæå®ããç¯å²ãéžè±ãããã·ã¹ãã ãå£ããããªæäœãé¿ããŠãããŸããã
Pricing
01 Pay-as-you-go ããŒã¯ã³ãã©ã³
Enterprise ãšã³ã¿ãŒãã©ã€ãº
For heavy, production workloads needing maximum reliability â consumption-based tokens are served at higher priority than monthly-plan tokens. æå€§éã®ä¿¡é Œæ§ãæ±ããããé«è² è·ã»æ¬çªã¯ãŒã¯ããŒãåããåŸé課éã®ããŒã¯ã³ã¯ãæé¡ãã©ã³ã®ããŒã¯ã³ããé«ãåªå 床ã§åŠçãããŸãã
Fugu
When 1 agent is active ãšãŒãžã§ã³ãã 1 ã€ã®å Žå
You pay only the standard rate for that specific underlying model. ãã®åºç€ã¢ãã«ã®æšæºã¬ãŒãã®ã¿ããæ¯æãããã ããŸãã
When multiple agents are active è€æ°ã®ãšãŒãžã§ã³ãã皌åããŠããå Žå
We never stack model fees; you are charged a single rate based on the top tier model involved. ã¢ãã«æéãç©ã¿äžããããšã¯ãããŸãããé¢äžããæäžäœã¢ãã«ã«åºã¥ãåäžã®ã¬ãŒãã§èª²éãããŸãã
Fugu Ultra
Fixed pricing for fugu-ultra-20260615 fugu-ultra-20260615 ã®æéïŒäžåŸïŒ
Input å ¥å
$5
$10 when context > 272K $10ïŒã³ã³ããã¹ã272Kè¶ ïŒ
Output åºå
$30
$45 when context > 272K $45ïŒã³ã³ããã¹ã272Kè¶ ïŒ
Cached input ãã£ãã·ã¥å ¥å
$0.50
$1.00 when context > 272K $1.00ïŒã³ã³ããã¹ã272Kè¶ ïŒ
02 Subscription Plan ãµãã¹ã¯ãªãã·ã§ã³ãã©ã³
Monthly æé¡
Best for individuals and everyday hands-on use. Every tier includes both Fugu and Fugu Ultra â upgrade when you need longer, heavier, or more frequent sessions. å人ãŠãŒã¶ãŒãæ¥åžžçãªãå©çšã«æé©ããã¹ãŠã®ãã©ã³ã§ Fugu ãš Fugu Ultra ã®äž¡æ¹ããå©çšããã ããŸããããé·æéã»é«è² è·ã»é«é »åºŠã®äœæ¥ãå¿ èŠãªå Žåã¯äžäœãã©ã³ãžã
Subscribe before the end of July 2026 to get a free second month at your initial subscription tier. 2026 幎 7 ææ« ãŸã§ã«ãç»é²ããã ããšã ãå å ¥ããã ãããã©ã³ ã® 2 ãæç®ãç¡æ ã§ãæäŸããŸãã
Standard
$20 /month /æ
Lightweight daily usage 軜éãªæ¥åžžå©çšã«
For occasional API calls, small experiments, and trying Fugu in personal workflows. äœé »åºŠã® API å©çšãå°èŠæš¡ãªå®éšãå人ã¯ãŒã¯ãããŒã§ã®è©Šçšã«ã
Baseline allowance æšæºã®å©çšæ
Pro
$100 /month /æ
Focused working sessions éäžããäœæ¥ã»ãã·ã§ã³ã«
For regular coding, review, research, and analysis sessions throughout the week. æ®æ®µã®ã³ãŒãã£ã³ã°ãã¬ãã¥ãŒã調æ»ãåæã»ãã·ã§ã³ã«ã
10à Standard usage Standard ã® 10 åã®å©çšæ
Max
$200 /month /æ
Heavy long-running workloads é·æéã®é«è² è·ã¯ãŒã¯ããŒãã«
For power users who keep Fugu active across deeper, longer-running tasks. ããæ·±ãé·æéã®ã¿ã¹ã¯ã§ Sakana Fugu ãç¶ç¶çã«äœ¿ããã¯ãŒãŠãŒã¶ãŒåãã
20à Standard usage Standard ã® 20 åã®å©çšæ
Start Using Sakana Fugu ä»ããã¯ããã
FAQ
Sakana Fugu is available through an OpenAI-compatible API. Point your existing client or coding harness at the Fugu endpoint with your API key and start sending requests â no SDK migration required. Sakana Fugu 㯠OpenAI äºæ API ãéããŠå©çšã§ããŸããæ¢åã®ã¯ã©ã€ã¢ã³ããã³ãŒãã£ã³ã°ããŒãã¹ããAPI ããŒãšãšãã« Fugu ã®ãšã³ããã€ã³ãã«åããŠãªã¯ãšã¹ããéãã ãã§ãââSDK ã®ç§»è¡ã¯å¿ èŠãããŸããã
Fugu balances latency and quality, making it a strong default for everyday coding and interactive work. Fugu Ultra prioritizes answer quality on complex, multi-step reasoning, coordinating more expert agents when accuracy and depth matter most, at the cost of response time. Early users reach for Fugu Ultra on demanding tasks like paper reproduction, Kaggle competitions, and paper or patent research. Fugu ã¯ã¬ã€ãã³ã·ãšå質ã®ãã©ã³ã¹ãåããŠãããæ¥åžžçãªã³ãŒãã£ã³ã°ãã€ã³ã¿ã©ã¯ãã£ããªäœæ¥ã«é©ããæšæºã¢ãã«ã§ããFugu Ultra ã¯ãè€éã§å€æ®µéã®æšè«ã«ãããŠåçå質ãæåªå ããç²ŸåºŠãšæ·±ããéèŠãªå Žé¢ã§ã¯ããå€ãã®å°éãšãŒãžã§ã³ãã飿ºãããŸãïŒãã®åãå¿çæéã¯é·ããªããŸãïŒãå è¡ãŠãŒã¶ãŒã¯ãè«æã®åçŸãKaggle ã³ã³ããã£ã·ã§ã³ãè«æã»ç¹èš±èª¿æ»ãªã©ã®é£ããã¿ã¹ã¯ã§ Fugu Ultra ãæŽ»çšããŠããŸãã
Fugu Ultra relies on the full agent pool to deliver its performance, so its pool is fixed. For Fugu, you can opt out of specific models from the settings menu on our console page to match your data, privacy, and compliance needs. Fugu Ultra ã¯ãã®æ§èœãçºæ®ããããã«å šãšãŒãžã§ã³ãããŒã«ãå©çšãããããããŒã«ã¯åºå®ã§ããFugu ã«ã€ããŠã¯ãã³ã³ãœãŒã«ããŒãžã®èšå®ã¡ãã¥ãŒããç¹å®ã®ã¢ãã«ãé€å€ã§ããããŒã¿ã»ãã©ã€ãã·ãŒã»ã³ã³ãã©ã€ã¢ã³ã¹ã®èŠä»¶ã«åãããããŸãã
We aim to give users the best performance available. When a new frontier model is released publicly, we expect to spend roughly two weeks training and evaluating updated Fugu models before rolling them out. ç§ãã¡ã¯å©çšå¯èœãªæé«ã®æ§èœããŠãŒã¶ãŒã«æäŸããããšãç®æããŠããŸããæ°ããããã³ãã£ã¢ã¢ãã«ãäžè¬å ¬éãããå Žåãã¢ããããŒãçã® Sakana Fugu ã¢ãã«ã®ãã¬ãŒãã³ã°ãšè©äŸ¡ã«çŽ 2 é±éãããããã®åŸé 次æäŸéå§ããŠããäºå®ã§ãã
We offer both subscription and pay-as-you-go plans, and every plan includes access to both Fugu and Fugu Ultra. The subscription plan has three monthly tiers: Standard ($20/month) is great for lightweight daily use; Pro ($100/month) provides 10à the usage of Standard, ideal for a few focused working sessions each week; and Max ($200/month) provides 20à the usage of Standard, built for heavy, long-running workloads. The pay-as-you-go plan bills by token usage instead of a monthly allowance, giving you elastic capacity for spikes and large jobs â ideal for enterprise customers. With it, Fugu is charged at the standard rate of the underlying model, and when multiple agents are active we never stack fees: you pay a single rate based on the top tier model involved. Fugu Ultra (fugu-ultra-20260615) is priced per 1M tokens at $5 input, $30 output, and $0.50 cached input, with higher rates ($10 / $45 / $1.00) for contexts above 272K tokens. ãµãã¹ã¯ãªãã·ã§ã³ãã©ã³ãšåŸé課éãã©ã³ã®äž¡æ¹ããçšæããŠããããã¹ãŠã®ãã©ã³ã§ Fugu ãš Fugu Ultra ã®äž¡æ¹ããå©çšããã ããŸãããµãã¹ã¯ãªãã·ã§ã³ãã©ã³ã«ã¯æé¡ 3 ã€ã®ãã©ã³ããããŸããStandardïŒ$20/æïŒã¯è»œéãªæ¥åžžå©çšã«æé©ãProïŒ$100/æïŒã¯ Standard ã® 10 åã®å©çšéãæäŸãé±ã«æ°åã®éäžäœæ¥ã«åããŠãããMaxïŒ$200/æïŒã¯ Standard ã® 20 åã®å©çšéãæäŸãé·æéã«ãããè² è·ã®é«ãã¯ãŒã¯ããŒãåãã§ããåŸé課éãã©ã³ã¯æé¡ã®å©çšæ ã§ã¯ãªãããŒã¯ã³äœ¿çšéã«å¿ããŠèª²éãããã¹ãã€ã¯ãå€§èŠæš¡ãžã§ãã«å¯Ÿããæè»ãªãã£ãã·ãã£ãæäŸããŸãââãšã³ã¿ãŒãã©ã€ãºã®ã客æ§ã«æé©ã§ãããã®ãã©ã³ã§ã¯ Fugu ã¯åºç€ã¢ãã«ã®æšæºã¬ãŒãã§èª²éãããè€æ°ã®ãšãŒãžã§ã³ãã皌åããŠããå Žåã§ãæéãç©ã¿äžããããšã¯ãããŸãããé¢äžããæäžäœã¢ãã«ã«åºã¥ãåäžã®ã¬ãŒãã§ãæ¯æãããã ããŸããFugu UltraïŒfugu-ultra-20260615ïŒã¯ 100 äžããŒã¯ã³ããããå ¥å $5ãåºå $30ããã£ãã·ã¥å ¥å $0.50 ã§ã272K ããŒã¯ã³ãè¶ ããã³ã³ããã¹ãã§ã¯ããé«ãã¬ãŒãïŒ$10 / $45 / $1.00ïŒãé©çšãããŸãã
Yes. Think of Fugu pricing as a single blended rate for the active agent pool, not a sum of every model used. If your pool contains only Model A, requests are billed at Model A's rate. If your pool contains Models A, B, and C, you still pay only one rate: the rate of the top tier model among A, B, and C. In other words, adding more agents does not multiply the bill; it only determines which single model rate applies to that configured pool. ã¯ãã Sakana Fugu ã®æéã¯ã䜿çšãããã¹ãŠã®ã¢ãã«ã®åèšã§ã¯ãªãã皌åäžã®ãšãŒãžã§ã³ãããŒã«ã«å¯Ÿããåäžã®ãã¬ã³ãã¬ãŒãã ãšèããŠãã ãããããŒã«ã«ã¢ãã« A ã ããå«ãŸããŠããå Žåããªã¯ãšã¹ãã¯ã¢ãã« A ã®ã¬ãŒãã§èª²éãããŸããããŒã«ã«ã¢ãã« Aã»Bã»C ãå«ãŸããŠããå Žåã§ãããæ¯æã㯠1 ã€ã®ã¬ãŒãââAã»Bã»C ã®ãã¡æäžäœã¢ãã«ã®ã¬ãŒãââã ãã§ããã€ãŸãããšãŒãžã§ã³ããå¢ãããŠãè«æ±ãç©ç®ãããããã§ã¯ãªãããã®æ§ææžã¿ããŒã«ã«é©çšãããåäžã®ã¢ãã«ã¬ãŒããæ±ºãŸãã ãã§ãã
Yes. Token usage and the corresponding cost are reported per request, so you can monitor spend in real time and forecast costs before scaling up. ã¯ããããŒã¯ã³äœ¿çšéãšå¯Ÿå¿ããã³ã¹ãã¯ãªã¯ãšã¹ãããšã«å ±åããããããæ¯åºããªã¢ã«ã¿ã€ã ã§ææ¡ããã¹ã±ãŒã«ã¢ããåã«ã³ã¹ããäºæž¬ããããšãã§ããŸãã
Usage data helps us keep improving Fugu's performance, and we're grateful when customers share it. That said, it's entirely your choice â you can opt out of training data usage at any time from our console page. åŠç¿ããŒã¿å©çšã«ã€ããŠã¯ã³ã³ãœãŒã«ããŒãžãããã€ã§ããªããã¢ãŠãããããšãã§ããŸããã客æ§ã®ã倿ã§ãå ±æããã ããå Žå㯠Sakana Fugu ã®æ§èœåäžã«åœ¹ç«ãŠãŸãã
No. The specific models Fugu selects and how it coordinates them are proprietary, so this routing information is not exposed by design. ãããã Sakana Fugu ãéžæããå ·äœçãªã¢ãã«ãããããã©ã®ããã«é£æºããããã¯ç¬èªæè¡ã§ããããã®èšèšæ å ±ã¯å ¬éããŠããŸããã
Yes, Fugu is available from outside Japan. However, we do not provide services to users in EU (European Union) or EEA (European Economic Area) member states (please refer to our Terms of Service for details). Additionally, in other regions, access may not be available due to network conditions or local regulations. ã¯ããæ¥æ¬åœå€ããããå©çšããã ããŸãããã ããEUïŒæ¬§å·é£åïŒããã³EEAïŒæ¬§å·çµæžé åïŒå çåœãžã®ãµãŒãã¹æäŸã¯è¡ã£ãŠãããŸããïŒè©³çްã¯å©çšèŠçŽãã確èªãã ããïŒããŸãããã以å€ã®å°åã«ãããŸããŠããéä¿¡ç°å¢ãçŸå°ã®åçš®èŠå¶çã«ãã£ãŠãå©çšããã ããªãå ŽåãããããŸãã
Contact Us
Get in touch to learn more about access, plans, and enterprise deployment. ã¢ã¯ã»ã¹æ¹æ³ããã©ã³ããšã³ã¿ãŒãã©ã€ãºåãã®å°å ¥ã«é¢ãã詳现ã«ã€ããŠã¯ããåãåãããã ããã