OpenRouter's Fusion Bets a Model Cocktail Can Outsmart Claude Fable 5 on a Budget 🧠

OpenRouter launched Fusion on June 13, 2026, an API that sends a single prompt to multiple AI models in parallel, then uses a judge model and a synthesizer to merge the results into one grounded answer. The product is positioned as offering "Fable-level intelligence at half the price," arriving one day after Anthropic suspended its Claude Fable 5 and Mythos 5 models for foreign nationals worldwide following a U.S. export control directive tied to a disputed jailbreak finding.

When a prompt hits Fusion, OpenRouter dispatches it to a panel of models with web search and bash tools attached. A judge model then extracts consensus points, contradictions and blind spots from each response before a synthesizer, Claude Opus 4.8 by default, writes the final answer grounded in that analysis. The entire pipeline runs server-side, and users can switch their model string to "openrouter/fusion" for a default panel, attach a fusion tool so an existing model can call it selectively, or assemble a custom panel in the Fusion chatroom without code.

Benchmark results posted by OpenRouter on Perplexity's DRACO test, built from real user deep research requests, show Fable 5 paired with OpenAI's GPT-5.5 and synthesized by Opus topping the chart at 69%, compared with solo Fable 5 at 65.3%, though seven of its 100 tasks never ran because its own content filters blocked them. A cheaper combination, Gemini 3 Flash with the open-source Chinese models Kimi K2.6 and DeepSeek V4 Pro fused and synthesized by Opus, hit 64.7%, beating solo GPT-5.5 at 60% and solo Opus 4.8 at 58.8% while landing within roughly a point of solo Fable 5 at about half the cost. Pairing Opus 4.8 with a separate instance of itself scored 65.5%, a 6.7-point jump over solo Opus, and OpenRouter attributes roughly three quarters of that lift to the synthesis step, with the rest coming from genuine model diversity.

OpenRouter acknowledged that giving the panel live web access allowed models to surface DRACO's own grading rubric in search results, a contamination risk the company described as coincidental rather than deliberate, and said a one-line configuration change excluding the benchmark's hosting domains from the search tools was applied before any numbers were published.

OpenRouter stated that Fusion is not a full Fable replacement, noting that DRACO skips long-horizon work where Fable reportedly still leads, and that for coding, Fusion is designed to operate as a tool a coding model calls selectively rather than a wholesale swap.