KandaQuantum Opens Limited Release Of Fuga v2 Ryusei
The release frames recursive self-improvement as an operational claim rather than a research demo, with the vendor reporting doubled daily agent counts and a rebuilt command foundation one month after v1.
Reporting from 1 source: ASCII.jp.
KandaQuantum Inc., based in Chiyoda-ku, Tokyo, began a limited release of Fuga v2 - Ryusei on September 23, 2026. The orchestrator commands AI agent swarms from planning through execution and integration. The company says it can now stably command more than 2,000 agents per day, double the figure from Fuga v1, after running a semi-automated RSI loop that finds and fixes the system's own bottlenecks.
The limited release covers model support for OpenAI GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, Anthropic Claude Opus 5.5, xAI Grok 4.7, and Google Gemini 3.8 flash. Fuga estimates the difficulty of each task and assigns a model whose capabilities match it, using higher-tier models for design and integration and high-speed, low-cost models for routine implementation and research.
Allocation is set by target ratios per vendor, with share for vendors removed from the pool redistributed to the rest. Within a single development request, models from multiple companies divide roles as a series running from planning to parallel implementation, integration, results report generation, independent inspection, and merge into the base branch.
- OpenAI GPT-6 Astra / GPT-6 Sol / GPT-6 Luna: Supported by v2 - Ryusei
- Anthropic Claude Opus 5.5: Supported by v2 - Ryusei
- xAI Grok 4.7: Supported by v2 - Ryusei
- Google Gemini 3.8 flash: Supported by v2 - Ryusei
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.