📉Open Models Drop to 3% of AI Budgets, Lowest Share Ever Measured

Open models are losing your cheap AI budget fast

Open-weight models are getting squeezed out of the value tier. The LLM Atlas, which maps price against capability across AI models, shows an open-weight model is the best buy at only 3% of budgets. In June that number was 41%. Why should you care? If your team routes requests by cost, your default model choice probably came from a June benchmark. OpenAI's September and October releases, GPT-6 Luna and GPT-6.1 Terra, now cover the budgets open models used to win. Anthropic's Claude Haiku 5.5 is also strong at the mid-tier. Closed labs can distill their frontier models and ship whole families, which open models can't easily match. The numbers: DeepSeek V4 Flash held nearly a third of budgets at its peak, with Zhipu's GLM-5.2 just above it. Today, most of the remaining cheap-end wins belong to Google's Gemma 4 E4B and Xiaomi's MiMo-V2.6-Flash. Open models have held at least 6% of budgets at every quarter end since 2024, so 3% is a new low. The catch? Those earlier peaks came from single releases that didn't last. Worth watching: whether the next open release resets the curve. Scores are as of October 31, 2026, and the Atlas updates as new results arrive.