Open models crowd frontier leaderboard

The real story is not that local models have won; it is that the performance gap is now narrow enough for developers to make deployment, privacy, latency, and cost first-class selection criteria.

• –Closed models still occupy the very top of the chart, with Claude, Gemini, OpenAI, and Kimi leading the highest scores.
• –Open-weight contenders like DeepSeek, Gemma, Nemotron, and gpt-oss make self-hosting viable for more serious workflows than hobby chatbots.
• –Benchmark screenshots should be treated as directional, not definitive; Artificial Analysis itself notes its Intelligence Index is a composite with use-case limits.
• –For builders, the practical shift is leverage: teams can now prototype locally, keep sensitive data on their own hardware, and only escalate to hosted frontier APIs when the task justifies it.