OpenRouter says Jev took 27% of its classification requests
Platform share reported by the platform, with no accuracy number attached.
OpenRouter posted that Jev took 27% of weekly request volume in its classification category, which it says is nearly twice the share of DeepSeek V4 Flash, the model that previously held the top spot in that category. OpenRouter's framing is that Jev is "quickly beocming the primary choice for classification requests" on the platform, typo included.
It took 27% of weekly request volume in this category, nearly 2x the share of DeepSeek V4 Flash which previously held the top spot
That is one number from one router, and it measures requests, not results. Share of traffic in a category tells you what developers pointed at the endpoint last week. It does not tell you whether the answers were right, whether the calls were cheap, or whether the teams sending them kept the traffic there after they looked at their own evals. Nothing in the post attaches an accuracy figure, a latency figure or a price to the 27%.
What else is in the source
Earlier in the week Diogo Almeida posted that Jev is the top model at 1k to 10k context on OpenRouter. Same platform, different cut of the same kind of data, and again a ranking rather than a measurement of output quality. The two posts are separate claims about traffic mix, not a benchmark and not a comparison anyone ran head to head.
It is worth being precise about the DeepSeek V4 Flash comparison, because it is the part most likely to be repeated wrong. OpenRouter says Jev's share is nearly 2x that of DeepSeek V4 Flash in the classification category and that DeepSeek V4 Flash was previously top. It does not give DeepSeek V4 Flash's current percentage, does not say how long Jev has held the lead, and does not break out how much of the category total the two models cover between them. If you want the remaining 73%, the post does not have it.
For people building on it
Router share is a demand signal, and demand signals move fast in both directions. The useful thing about this number is capacity planning and the fact that a lot of other people's classification traffic is now sitting on the same model you might be about to put in a hot path. The thing it is not is evidence that Jev is the right classifier for your labels. For that, the more relevant reading on this site is the work where someone measured output rather than volume, such as OpenRouter's own judging eval, and the cases where Jev came up short on specific tasks. A model can be the most requested option in a category and still be the wrong call on your data, and only your eval closes that gap.
jev is the top model at 1k-10k context on openrouter!