OpenRouter Launches Fusion AI API Amid Fable 5 Suspension

OpenRouter launched its Fusion API on June 12, 2026, designed to merge the responses from several AI models, including a judge model and a synthesizer, to provide high-quality answers. The Fusion approach aims to deliver capabilities similar to that of Claude Fable 5 while reducing costs significantly.

The introduction of Fusion comes at a strategic moment, shortly after Anthropic's Fable 5 and Mythos 5 faced suspension due to a U.S. export control directive linked to a controversial jailbreak issue. OpenRouter seized the opportunity to position Fusion as a cost-effective alternative, claiming it achieves "Fable-level intelligence at half the price."

Fusion operates by parallel processing prompts to multiple models, leveraging web search and bash tools for enhanced accuracy. A judge model evaluates responses, identifying consensus and discrepancies, followed by a synthesizer—typically Claude Opus 4.8—that provides a final, cohesive answer.

In testing against Perplexity's DRACO benchmark, Fusion's ensemble of AI models delivered results within 1% of the more expensive Fable 5, despite Fable's own performance being affected due to its content filtering blocking some tasks. OpenRouter's model combination is becoming increasingly relevant as restrictions around large language models (LLMs) become more commonplace.

Looking forward, industry observers will be monitoring how Fusion and similar approaches evolve in light of regulatory challenges, alongside the broader implications for AI accessibility and competition in the market.