Jlabs DigitalOCULUS
Jlabs Digital
RSSjlabsdigital.com
© 2026 Jlabs Digital. Not financial advice.
macro6 min read

OpenRouter might have undercut its own pitch

June 15, 2026
OpenRouterFusionAILLMcompound-modelssynthesisbenchmarkDRACOClaude Fable 5opinion

OpenRouter launched Fusion this week. The headline is that it matches Claude Fable 5 on a research benchmark for about half the cost, and they're calling it the smartest compound model on the market.

I read the whole thread, and the thing that stuck with me wasn't the headline. It was a number they dropped halfway down. They say roughly three quarters of Fusion's improvement comes from the synthesis step, and only about a quarter from using a mix of different models.

Cost vs Fable 5~½for near-equal quality
Quality gap to Fable 5~1%budget panel, fused
Gain from synthesis75%OpenRouter's own split
Gain from diversity25%the panel of models
Fusion · Figure 1 — Source: OpenRouter, company benchmarks (DRACO)
FUSION · FIGURE 1 Where the gains actually come from OpenRouter's own breakdown of what makes Fusion smarter. 75% 25% SYNTHESIS The judge and synthesizer reconciling every model's answer into one. DIVERSITY Using a panel of different models. Source: OpenRouter, company benchmarks (DRACO) Opinion — JLabs
By OpenRouter's own split, synthesis — the judge and synthesizer reconciling answers — does three quarters of the work. The panel of models accounts for the other quarter.

That's a strange thing for OpenRouter to admit. Their business is selling access to lots of models. And here they are telling you the models are the minor ingredient.

01

How it actually works

Fusion announcement

You send a prompt. Fusion sends it out to several models at once. A “judge” model reads all the answers and works out where they agree, where they contradict each other, and what each one caught that the others missed. Then a “synthesizer” model writes the final answer off the back of that.

Fusion · Figure 2 — Source: OpenRouter, Fusion announcement
FUSION · FIGURE 2 How Fusion answers one prompt Your question is fanned out, judged, then rewritten as a single answer. PROMPT your question PANEL OF MODELS A B C run in parallel · web + bash JUDGE maps agreement, conflicts, gaps and unique insights SYNTHESIZER writes the final answer from that map ANSWER one reply By OpenRouter's measurement, the judge and synthesizer steps do most of the work, not the panel. Source: OpenRouter, Fusion announcement Opinion — JLabs
One prompt is fanned out to a panel, judged for agreement and conflict, then rewritten as a single reply.

So the part that moves the needle, by their own measurement, is the judging and the writing-up at the end. Not the panel itself. I find that genuinely interesting, and a little awkward for them.

02

The cheap-models result

DRACO (Perplexity)

This is the part I keep coming back to. OpenRouter says a panel of budget models (Gemini 3 Flash, Kimi K2.6, and DeepSeek V4 Pro) came within about 1% of Fable 5 once they were fused together. That same panel apparently beat solo GPT-5.5 and solo Opus 4.8 outright.

Fusion · Figure 3 — Source: OpenRouter benchmarks · DRACO (Perplexity)
FUSION · FIGURE 3 Cheap models, fused, kept up Relative results OpenRouter reported on the DRACO deep-research benchmark. QUALITY (DRACO) Claude Fable 5 single frontier model benchmark top Budget panel, fused Gemini 3 Flash · Kimi K2.6 · DeepSeek V4 Pro within ~1% Solo GPT-5.5 single frontier model beaten Solo Opus 4.8 single frontier model beaten COST ½ the price of Fable 5, for near-equal quality. Bars are relative, not exact: OpenRouter didn't publish per-model scores, only these comparisons. Source: OpenRouter benchmarks · DRACO (Perplexity) Opinion — JLabs
A fused panel of cheap models landed within ~1% of Fable 5 and beat two solo frontier models — at roughly half the cost. Bars are relative; OpenRouter did not publish per-model scores.

If that holds up, it suggests raw model quality matters less than I'd assumed. You might not need the best model in the room. You need a few decent ones and a good way to combine what they say.

I'm hedging on “if it holds up” for a reason. This is one benchmark, run by the company selling the thing.

03

Where I get skeptical

Two things bug me.

The price claim. “Half the cost of Fable” reads clean, but you're not running one model anymore. You're running a whole panel of them, plus a judge, plus a synthesizer, on every single query. Cheaper than Fable for comparable quality on this test, maybe. It's still a lot more compute and a lot more waiting than a single model call, and the “just call one slug” framing skips past that.

The benchmark. They tested on DRACO, a deep-research benchmark from Perplexity. Deep research happens to be close to the best possible case for this approach, because the actual work is reconciling a pile of sources. I'd want to see how Fusion does on quick factual questions, on coding, on anything creative, before I bought “smartest compound model on the market” as a general statement. On a one-line question, fanning out to five models and running a judge over them is probably just slower and pricier for no real gain.

04

What I take from it

For the last couple of years the conversation has been about the model. Which one scored highest, which lab shipped what. Fusion's own data points somewhere else, up a layer, into how you route a question and combine the answers you get back.

I don't think this means models stop mattering. A panel of bad models won't save you. But OpenRouter set out to sell a smarter model and ended up making a fairly good case that the answer matters less than what you do with it afterward. I doubt that was the plan.

Opinion — Jlabs Research. Sources: OpenRouter Fusion announcement and company benchmarks; DRACO deep-research benchmark (Perplexity). Figures are relative illustrations; OpenRouter did not publish per-model scores. Not investment advice.

Share this report

Related Posts