OpenRouter has launched Fusion, a new API that combines multiple AI models to deliver high-quality outputs at reduced costs, following Anthropic's recent model suspension.
Summary
- OpenRouter has launched Fusion on June 12, a new server-side API that utilizes multiple AI models to provide high-quality outputs.
- This cost-effective solution performs nearly as well as Fable 5 on Perplexity's DRACO benchmark while being approximately 50% cheaper.
- The launch comes shortly after Anthropic paused Fable 5 and Mythos 5 due to U.S. export controls.
OpenRouter has introduced Fusion, a versatile API designed to combine the strengths of various AI models to match the performance of high-end models like Claude Fable 5.
Fusion operates by simultaneously querying multiple models and employing a judge and synthesizer to produce a comprehensive response.
The timing of the launch is critical, following the suspension of Anthropic’s Fable 5 and Mythos 5 due to regulatory issues, which has opened the door for alternatives. OpenRouter promoted Fusion on X, highlighting its ability to deliver "Fable-level intelligence at half the cost."
Introducing the Fusion API, the leading compound model available.
Fusion delivers Fable-level intelligence at half the cost.
How it works 👇 pic.twitter.com/OTUQAdTQjU
— OpenRouter (@OpenRouter) June 13, 2026
Affordable Performance with Fusion
When a prompt is sent to Fusion, it queries a panel of models, each equipped with web search and bash tools. A judge model assesses the responses for consensus and discrepancies, while a synthesizer—often Claude Opus 4.8—compiles the final output.
This entire process is executed server-side, allowing users to easily configure their settings without coding. OpenRouter tested Fusion using DRACO, Perplexity’s benchmark based on real user queries, where the combination of Fable 5 and GPT-5.5 achieved a score of 69%. In contrast, Fable scored 65.3%, with some tasks blocked due to content filters.
OpenRouter emphasizes the affordability of its solution, as demonstrated by the Gemini 3 Flash and open-source models, which collectively scored 64.7%—surpassing GPT-5.5 and Opus 4.8 while costing around half as much.
Even a pairing of Opus 4.8 with itself achieved a score of 65.5%, with OpenRouter noting that about 75% of the improvement stemmed from the synthesis process.
However, there are concerns about the potential for models to inadvertently reveal DRACO's grading criteria through web access. OpenRouter has taken steps to mitigate this by excluding certain domains from search results.
Evaluating Fusion's Efficacy
While Fusion provides a cost-effective alternative, OpenRouter clarifies that it should not be seen as a complete replacement for Fable. The DRACO benchmark does not account for long-term tasks, where Fable may still excel. Fusion is intended as a supplementary tool for scenarios requiring multiple perspectives and cross-checking.
For complex research and planning, Fusion appears to enhance outcomes by leveraging a collaborative approach, demonstrating that a combination of accessible models can achieve similar results to more expensive standalone options.
The launch has been met with a favorable response, with AI researcher Andrew Trask calling it "a way bigger deal than it seems," suggesting it could reduce the dominance of frontier labs in AI technology. However, some skeptics have raised concerns about coding results and transparency in comparison to Fable 5.
