Sakana AI Releases Fugu Ultra: Multi-Agent Orchestration System with 1M Context Window at $5/$30 per Million Tokens
Sakana AI has released Fugu Ultra, a multi-agent orchestration system that routes tasks across pools of underlying models rather than operating as a single monolithic model. The system supports a 1M token context window and is priced at $5 per million input tokens and $30 per million output tokens.
Fugu Ultra at a glance
Sakana AI Releases Fugu Ultra: Multi-Agent Orchestration System with 1M Context Window
Sakana AI has released Fugu Ultra, the higher-performance model in its Fugu family, now available through OpenRouter. Unlike traditional language models, Fugu Ultra is a learned multi-agent orchestration system trained to route tasks across a swappable pool of underlying models and recursively call instances of itself.
Architecture and Capabilities
According to Sakana AI, Fugu Ultra prioritizes answer quality on complex, multi-step reasoning, coding, and agentic workflows. The system supports:
- 1 million token context window
- Configurable reasoning effort
- Tool calling
- Built-in web search capabilities
Orchestration tokens consumed by the system are billed as standard input/output tokens, with no separate pricing tier for the routing logic.
Pricing
Fugu Ultra is priced at:
- Input: $5 per million tokens
- Output: $30 per million tokens
The effective price can be 60-80% lower when prompt caching is applied for repeated context, according to OpenRouter's monitoring data.
Technical Approach
Rather than training a single large model, Sakana AI's approach involves training a language model to act as an orchestrator, deciding which underlying models to route tasks to and when to recursively invoke additional instances of itself. This architecture represents a departure from the monolithic model paradigm adopted by most major AI labs.
The model was released June 24, 2026, according to OpenRouter's listing. Sakana AI has not disclosed benchmark scores, parameter count, or details about the underlying model pool at this time.
Availability
Fugu Ultra is currently available exclusively through OpenRouter, which forwards requests directly to Sakana AI's infrastructure with no intermediate routing layer.
What This Means
Sakana AI's orchestration approach represents a significant architectural departure from the single-model paradigm. By routing tasks to specialized models rather than attempting to embed all capabilities in one system, Fugu Ultra could potentially offer better cost-performance ratios for complex workflows. However, without published benchmarks, it's unclear how the system compares to frontier models like Claude 3.5 Sonnet or GPT-4 on standardized reasoning and coding tasks. The success of this approach will depend on whether the orchestration overhead is offset by more efficient task routing.
Related Articles
Google releases Gemini 4 Neon at $3/$15 per 1M tokens with a 1M-token context window
Google released Gemini 4 Neon, available today through the API at $3 per million input tokens and $15 per million output tokens. It is the fourth new Gemini 4 model in four weeks, with a context window of about one million tokens.
Tencent releases Hy4 flash: 220B MoE, 12B active parameters, 1M-token context
Tencent has released Hy4 flash, a multimodal Mixture-of-Experts model with 220B total parameters, 12B active per token, and a 1,048,576-token context window. Tencent calls it its most cost-effective Hy model to date and claims a per-task cost of 1/20 that of Hy4 on its internal WorkIndex benchmark.
SID AI releases SID-2 agentic search model; 12-agent Swarm scores 0.720 NDCG
SID AI has released SID-2, an agentic search model trained with reinforcement learning. A single agent reaches 0.585 NDCG in 3.11 seconds, and a 12-agent SID-2 Swarm scores 0.720 NDCG, which SID says is above GPT-6 Astra on its benchmark.
OpenAI releases GPT-Live-1.1 in the API, with standby mode billing idle WebRTC time at $0.02 per minute
OpenAI has released GPT-Live-1.1 in the API. The full-duplex voice model improves multilingual accents and transcription, and a new standby mode bills idle WebRTC time at $0.02 per minute instead of $0.05.
Comments
Loading...