Village's LLM Atlas maps the price-performance frontier, showing closed models now hold nearly every budget. Open-weight models are the best buy at only 3% of budgets, dominated by Google's Gemma 4 E4B and Xiaomi's Mi Mo-V2.6-Flash. In June, open models held 41% of the market, but Two Open AI releases, GPT-6 Luna and GPT-6.1 Terra, reclaimed significant ground in September and October.
Open models have historically led on value in bursts, typically following strong cheap releases from Deep Seek, but these gains rarely last as closed labs quickly undercut pricing. The latest closed releases are both cheaper and stronger than open alternatives, shrinking the value band to the bottom of the price range. Three factors explain this trend: closed labs can distill cheaper tiers from their own frontier models, they release whole model families covering multiple price points, and they benefit from massive inference scale on proprietary hardware.
This study uses Village's LLM Atlas, estimating capability scores from public benchmarks and cost per task from evaluator runs, with data current as of October 31, 2026.
Source: Hacker News · Summarized by HeadlinesBriefing