Physix FrontierPhysix Frontier · News Briefing Card (GeekPark · Oct 4, 2026)
Shanghai startup StartLux open-sources decision model, claims it beats Jev
KEY FACTS
- On September 15, TypeSafe AI released Jev, a decision model that only makes judgments.
- On September 30, Shanghai-based company StartLux open-sourced StartLux-Decision, claiming its benchmark scores beat Jev.
- Within two weeks, OpenAI, Cloudflare, Amazon, and Shanghai AI Lab successively launched similar models.
- StartLux reported that its 27B version scored higher than Jev on 31 of 38 tests in the Decision Index.
- StartLux was founded by Chen Danian, co-founder of Shanda, and focuses on open-source local deployment.
KEY DATA
$0.042 per million tokensJev input price
5 tiersStartLux-Decision size tiers
63.88 vs 57.91Decision Index composite score
about 2.71GB4B version Q4 quantized size
PHYSIX OBSERVATION
Decision models went from a debut to an all-out brawl within two weeks, showing that the technical barrier in the judgment layer is not high and that the real moat lies in local deployment and product assembly. Chinese teams collectively choosing open source both sidesteps Jev's closed-source restrictions and turns the Qwen base into a globally shared backbone. For users this is good news: the cost of high-frequency judgment will fall rapidly.
Source: GeekPark report