Japan quite dropped the model

Japan sakana dropped a model outperform Chat GPT on benchmarks It's called built by Sakana AI like Gemini Fugu Fugu GPT OPENAI Fugu coordinate models Fugu isn't a model like ChatGPT, Claude, or Gemini— it's the conductor that coordinates between them. FUGU MODEL ORCHESTRATION Sakana Fugu LLM Pool for each the task in parallel combines into one answer NO CONDUCTING REQUIRED just go direct SIMPLE TASK GPT Claude ANTHROPIC STRAIGHT TO THE MODEL For simple tasks, you'd still just use Claude or GPT directly. Fugu Orchestrating 12% OVERALL COMPLETE Deep research agent RUNNING Code review agent RUNNING Security audit agent DONE ACTIVITY LOG 00:00 research queued: deep research task 00:00 code queued: code review task 00:00 security queued: security audit task 00:01 research fetching sources for query set Fugu Orchestrating 23% OVERALL COMPLETE Deep research agent RUNNING Code review agent RUNNING Security audit agent DONE ACTIVITY LOG 00:00 research queued: deep research task 00:00 code queued: code review task 00:00 security queued: security audit task 00:01 research fetching sources for query set Fugu Orchestrating 37% OVERALL COMPLETE Deep research agent RUNNING Code review agent RUNNING Security audit agent DONE ACTIVITY LOG 00:00 research queued: deep research task 00:00 code queued: code review task 00:00 security queued: security audit task 00:01 research fetching sources for query set 00:03 research scanning 34 candidate documents Fugu Orchestrating 49% OVERALL COMPLETE Deep research agent RUNNING Code review agent RUNNING Security audit agent RUNNING ACTIVITY LOG 00:00 research queued: deep research task 00:00 code queued: code review task 00:00 security queued: security audit task 00:01 research fetching sources for query set 00:03 research scanning 34 candidate documents 00:05 security loading auth flow graph Fugu Orchestrating 59% OVERALL COMPLETE Deep research agent RUNNING Code review agent RUNNING Security audit agent RUNNING ACTIVITY LOG 00:00 research queued: deep research task 00:00 code queued: code review task 00:00 security queued: security audit task 00:01 research fetching sources for query set 00:03 research scanning 34 candidate documents 00:05 security loading auth flow graph 00:05 research cross-referencing citations 00:05 code scanning diff, 12 files changed FUGU GPT-4.5 reportedly outperforming GPT 5.5 On SWE SWE-BENCH PRO - RESOLVE RATE (%) Fugu Ultra 73.7 Opus 4.8 69.2 GPT-5.5 58.6 Gemini 3.1 Pro 54.2 The AI Orchestra Hover over any model. Watch Fugu route the right instrument to the right note. FUGU CONDUCTOR ChatGPT Conversational Claude Reasoning Gemini Multimodal Llama Open Source Mistral Fun Grok Near-line Perplexity Research DeepSeek Deep Reasoning So collective of world's working