🧠Tencent Hy4 Ships With Vision and Hits 96.15% WorkBuddy Pass Rate

Your open-source model shortlist just got a new contender

Tencent just dropped Hy4, and it's the first model from them that can actually look at images. It also landed at the top of an open-source financial analysis blind test, beating DeepSeek, Kimi, and GLM. Open-source models have been trailing on the messy, high-stakes stuff, so this one is worth your attention. If you're building agents or internal tools, the hallucination numbers are the part to read twice. Hy4 cut its hallucination rate from 12.95% to 7.2%, and errors from unsupported assumptions dropped from 23.1% to 5.6%. That's the difference between a demo and something you'd let near a spreadsheet. It's also tuned for chart analysis, CAD and Blender-style work, and multi-step business logic, so frontend and backend orchestration tasks got attention too. The raw numbers: 3.04 on the financial evaluation across 579 real tasks, versus 2.97 for DeepSeek-V4.1-Flash-1017 and 2.68 for both Kimi K3.1 Pro and GLM-5.3. WorkBuddy scored an average of 83.18 with a 96.15% pass rate. Game dev benchmarks rose 19.1% on GameDev and 9.8% on GameLogic. The API is already reachable through OpenRouter, Hermes, and OpenCode. Watch the weak spot: Tencent admits open-ended experimental design lags frontier models. Benchmarks are Tencent's own, so run your evals before you bet a product on it.