OpenRouter Releases Fusion System Combining Multiple AI Models to Surpass Claude Fable 5
Fusion shows that combining existing models can match or beat a frontier model's performance, potentially shifting the AI market toward orchestration rather than single-model supremacy.
Reporting from 1 sources: GIGAZINE.
OpenRouter has released a system called Fusion that runs multiple AI models in parallel, combining their outputs to achieve benchmark scores higher than Anthropic's top model Claude Fable 5. Combinations like Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro exceeded Fable 5's scores. A cheaper mix of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro came within 1% of Fable 5 at half the cost. A bug where Claude Opus 4.8 was used as the review model for certain combinations was acknowledged and promised to be fixed.
OpenRouter's Fusion system sends a user's input to multiple models in parallel, then uses a review model to analyze each response and a calling model to generate the final answer. The company published benchmark graphs showing that combinations such as Claude Opus 4.8 plus GPT-5.5, or Claude Opus 4.8 plus GPT-5.5 plus Gemini 3.1 Pro, scored higher than Claude Fable 5, which Anthropic calls its best model ever. A lower-cost combination of Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro achieved a score within 1 percent of Fable 5 at half the cost.
Shortly after Fusion's release on June 14, 2026, a user reported that specifying DeepSeek V4 Pro, DeepSeek V4 Flash, and Kimi K2.6 caused Claude Opus 4.8 to be used as the review model. OpenRouter developer Toven acknowledged the issue as a bug and promised a fix. Fusion is available to try through an OpenRouter account.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.