OpenRouter launches Fusion API: "Putting a good mold" to achieve Fable level 5 intelligence at half the cost
OpenRouter launches the Fusion API multi-model collaboration system. The Gemini 3 Flash+Kimi K2.6+DeepSeek V4 Pro combination achieves close performance at about half the cost of Claude Fable 5. The number of OpenRouter access models exceeds 400.
OpenRouter launches Fusion API: multi-model "put together a good model" to challenge Fable 5 at half the cost
OpenRouter, the world's largest AI large model API aggregation platform, launched Fusion API on June 14 - a multi-model collaboration system that sends user input to multiple models at the same time, and the review model generates a unified answer after aggregation and analysis. Judging from benchmark scores, some multi-model combinations have surpassed Anthropic's flagship Claude Fable 5.
Fusion's working link is divided into three steps: the first step is to send requests in parallel and let multiple models answer their own responses; the second step is for the review model to read the answers and output structured analysis; the third step is to call the model to generate the final answer based on the analysis content.
Full ranking on SWE-Bench:
| Type | Model Combination | Score |
|---|---|---|
| Fusion | Fable 5 + GPT-5.5 | 69.0% |
| Fusion | Opus 4.8 + GPT-5.5 + Gemini 3.1 Pro | 68.3% |
| Fusion | Opus 4.8 + GPT-5.5 | 67.6% |
| Fusion | Dual Opus 4.8 | 65.5% |
| Solo | Claude Fable 5 | 65.3% |
| Fusion | Gemini 3 Flash + Kimi K2.6 + DeepSeek V4 Pro | 64.7% |
| Solo | DeepSeek V4 Pro | 60.3% |
| Solo | GPT-5.5 | 60.0% |
| Solo | Claude Opus 4.8 | 58.8% |
| Solo | Kimi K2.6 | 53.7% |
| Solo | Gemini 3.1 Pro | 45.4% |
| Solo | Gemini 3 Flash | 43.1% |
The most notable is the "affordable combo" - Gemini 3 Flash + Kimi K2.6 + DeepSeek V4 Pro - which manages a benchmark score difference of just 1% (64.7% vs 65.3%) at about half the cost of the Claude Fable 5. What’s more noteworthy is that the three models in this combination do not have high individual running scores (43.1%, 53.7%, 60.3%), but through Fusion’s review-synthesis mechanism, the synergistic effect greatly exceeds their respective upper limits.
There are also two counter-intuitive findings: first, the Fable 5 + GPT-5.5 combination (69.0%) not only surpasses the Fable 5 single running score (65.3%), but also surpasses the GPT-5.5 single running score of about 60.0%, indicating that even the strongest model can obtain additional gains through multi-model collaboration; second, the dual Opus 4.8 combination (65.5%) actually slightly surpasses Fable 5. The separate running score (65.3%) means that the collaboration of two sub-strong models can surpass the single strongest model in certain scenarios.
OpenRouter also announced that the number of access models exceeded 400. The essence of Fusion API is "exchanging cheap computing power for intelligence" - by calling multiple low-priced models in parallel and using a review model to make a decision, it opens up a new path to bypass single vendor lock-in and obtain flagship AI capabilities at a lower cost.
Reviews