OpenRouter launched Fusion on June 12, a server-side API that distributes prompts across multiple AI models simultaneously and synthesizes their responses into a unified answer. The company developed the system to deliver performance comparable to Claude Fable 5 at approximately half the cost, according to benchmark testing on Perplexity's DRACO dataset. The launch occurred shortly after Anthropic suspended access to Fable 5 and Mythos 5 following a U.S. export control directive that cited a disputed jailbreak finding, creating a market gap for high-performance AI alternatives accessible to international users.
OpenRouter Fusion Uses Multi-Model Panel Architecture
Fusion operates by sending a single prompt to multiple AI models in parallel, with each model receiving access to web search and bash tools. A judge model then extracts consensus points, contradictions, and blind spots from all responses. Claude Opus 4.8 serves as the default synthesizer, which produces the final answer based on the judge's analysis. Users can access Fusion by switching their model string to "openrouter/fusion" for a default panel configuration, adding a fusion tool for selective calls, or building custom panels through the Fusion chatroom interface without coding.
Budget AI Panel Scores 64.7% on DRACO Benchmark
OpenRouter tested Fusion on DRACO, Perplexity's benchmark constructed from real user deep research requests. A panel combining Fable 5 and OpenAI's GPT-5.5, synthesized by Opus, achieved the highest score at 69%. Solo Fable 5 scored 65.3%, though seven of its 100 tasks did not run due to content filter blocks. The budget configuration—Gemini 3 Flash combined with Kimi K2.6 and DeepSeek V4 Pro, synthesized by Opus—scored 64.7%, surpassing solo GPT-5.5 (60%) and solo Opus 4.8 (58.8%) at roughly half the cost of Fable 5. A panel pairing Opus 4.8 with a separate instance of itself scored 65.5%, representing a 6.7-point improvement over solo Opus; OpenRouter attributed approximately three-quarters of this gain to the synthesis step and the remainder to model diversity. The company identified a contamination risk where models accessed DRACO's grading rubric through web search results, which OpenRouter addressed by excluding the benchmark's hosting domains from search tools. All published scores reflect this corrected configuration.
Fusion Launch Follows Fable 5 Export Suspension
Anthropic suspended Fable 5 and Mythos 5 last week for all foreign nationals worldwide after a U.S. export control directive cited a disputed jailbreak finding. OpenRouter announced Fusion the next day, stating the API delivers "Fable-level intelligence at half the price." Fusion operates entirely on models routed through OpenRouter's infrastructure and does not resolve the underlying export control restrictions.
OpenRouter Identifies Fusion Limitations for Coding Tasks
OpenRouter stated that Fusion does not serve as a complete Fable 5 replacement. The DRACO benchmark excludes long-horizon work, where Fable 5 reportedly maintains an advantage. For coding applications, Fusion functions as a tool that coding models call selectively rather than a wholesale replacement. OpenRouter positioned Fusion for deep research, complex planning, and tasks where cross-checking multiple model perspectives provides value. AI researcher Andrew Trask called the launch "a way bigger deal than it seems," arguing that frontier labs will no longer exclusively control the frontier. Critics cited poor coding results, inadequate tool calling, and lack of transparency due to Fable 5's unavailability for direct comparison.
FAQ
What did OpenRouter launch on June 12?
OpenRouter launched Fusion, a server-side API that sends prompts to multiple AI models in parallel, uses a judge model to analyze responses, and employs a synthesizer to produce a unified answer.
How did the budget AI panel perform on the DRACO benchmark?
The budget panel combining Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro scored 64.7% on DRACO, landing within one percentage point of solo Fable 5's 65.3% at approximately half the cost.
Why did Anthropic suspend Fable 5 and Mythos 5?
Anthropic suspended Fable 5 and Mythos 5 last week following a U.S. export control directive that cited a disputed jailbreak finding, affecting all foreign nationals worldwide.
