Sakana Marlin
#94 v Frontier jazykové modelysakana-ai · od 2026-06-15 · 5× · naposledy 29. 6. 2026
Sakana Marlin is the first commercial product from Tokyo-based AI startup Sakana AI, launched on June 15, 2026. It is not a standalone large language model but an autonomous B2B research agent that orchestrates multiple frontier LLMs (including OpenAI o4-mini, Google Gemini 2.5 Pro, and DeepSeek-R1) via the company's proprietary AB-MCTS algorithm (Adaptive Branching Monte Carlo Tree Search, NeurIPS 2025 Spotlight). The agent operates without human intervention for up to eight hours, delivering strategy reports of up to 100 pages with executive summary slides. Its target audience is financial institutions, corporate strategy teams, consulting firms, and think tanks, where it is positioned as a "Virtual CSO" (Chief Strategy Officer).
Vlastnosti
| Coding Benchmark (SWE-bench Pro %) | No value available – Marlin is not an LLM and does not publish scores on standard coding benchmarks like SWE-bench Pro |
| Context Window (Tokens) | Not published – Marlin is an agent framework without its own context window; the underlying base LLMs used are not specified |
| Multimodal (Image/Video) | Not documented – Marlin is a text-based strategy research agent; no official information found on image or video processing as input |
| Price per 1M Input Tokens | Not applicable / not published – billing is per credit/run, not per token |
| Price Tier (Monthly) | Pay-as-you-go: 100 credits/run at ¥98/credit (~$66/run) | Pro: ¥150,000/month (~$1,000) with 2,000 credits | Team: ¥400,000/month (~$2,700) with 6,000 credits | Enterprise: custom |
| Context Window | Not applicable / not published – Marlin is a SaaS agent without its own context window; per the vendor, a single run comprises hundreds to thousands of LLM requests over up to 8 hours |
| Memory Requirement | Not applicable – fully cloud-hosted SaaS service; no local operation, no published hardware requirements |