Fugu Max
Last verified September 12, 2026 · Sakana AI pricing ↗$2.00/1M input · $6.00/1M output · 1.0M context · Sakana AI
Count Tokens
—
Tokens
—
Input Cost
—
Output Cost
Estimate Monthly Cost
Monthly Cost Estimator
Quick:
<$0.0001/mo
Pick a preset above or enter custom usage
Alternatives to Fugu Max
Pricing Details
Sakana AIORCHESTRATOR, NOT A SINGLE MODEL: Sakana describes Fugu as 'a language model trained to call various LLMs in an agent pool, including instances of itself recursively'. Model id fugu-max resolves to fugu-max-v1.0 (released 2026-09-11; OpenRouter listed sakana/fugu-max at 05:32 UTC that day). Card per console.sakana.ai/pricing: $2.00 input, $6.00 output, $0.25 cached input, 'regardless of context length' (no >272K tier, unlike Fugu Ultra). No batch rate, no cache-write charge listed, no published rate limits. web_search and web_fetch $0.007 per call on Sakana ($0.01 per call on OpenRouter's listing), and the models page warns 'a single query may require multiple calls'. THE STORED PRICE IS A FLOOR: the pricing page states orchestration tokens 'represent real token usage outside of the input and output tokens and will be counted in the final price of the request. The price will be the same as standard input and output tokens.' They are returned in input_tokens_details.orchestration_input_tokens / orchestration_input_cached_tokens and output_tokens_details.orchestration_output_tokens and are ADDITIVE to input_tokens and output_tokens (total_tokens = all five), unlike OpenAI's reasoning_tokens which are a subset. Sakana publishes no typical orchestration share; FAQ Q9 says routing is 'not exposed by design'. The usage paragraph names Fugu Ultra (models page: Ultra and Cyber), but Sakana's own OpenRouter description for fugu-max says 'Orchestration tokens consumed by the system are billed as standard input/output tokens'. Break-even orchestration overhead vs Sonnet 5 ($2/$10): 11.1% on a 30K-in/2K-out turn, 62.5% on 1K-in/5K-out; vs Terra ($2/$12) 16.7% / 93.8%; vs Kimi K3 ($3/$15) 66.7% / 143.8%. IDENTICAL CARD TO QWEN3.8 MAX on all three lines ($2.00/$6.00/$0.25). Pool: 'our largest pool of open-weights and specialized models, including NVIDIA Nemotron family'; the pool diagram labels it 'Closed & open models' plus Sakana Namazu; composition not disclosed; fixed pool, no opt-out (base Fugu tier allows opt-out and bills at the underlying model's rate). Sakana-reported benchmarks (2026-09-11 chart, baselines' provenance unstated): Terminal Bench 2.1 89.5, GPQA-D 95.5, AA-LCR 83.0, GDP.pdf 26.0, AutomationBench 49.5, SWEFish 62.5, HLE text 44.7, DeepSWE 70.8, CharXiv 88.1, Chartography 37.0. The chart plots Gemini 3.8 Flash at its 2027 price ($1.50/$7.50) rather than the $0.75/$3.75 in force through 2026-12-31, and DeepSeek V4 Pro at its peak rate. NO ARTIFICIAL ANALYSIS ENTRY as of 2026-09-12 (/models/fugu-max 404). OpenRouter 2026-09-12: 72 tok/s P50, 8.39 s TTFT, single provider; DataNorth read 13 tok/s and 5.37 s hours after launch. Reasoning effort high/xhigh (max maps to xhigh). Chat Completions, Responses and Anthropic Messages APIs at api.sakana.ai/v1. Not available in the EU/EEA. Included in the $20/$100/$200 subscriptions. Estimated tokens (undisclosed tokenizer).
Input / 1M tokens
$2
Output / 1M tokens
$6
Context Window
1.0M
Max Output
128K
Price History
Launched at current price on 2026-09-11. No price changes recorded since.
Frequently Asked Questions
Common questions about Fugu Max pricing and usage