Epoch AI Measures Open-Weight Models Four Months Behind Closed Frontier

Epoch AI's analysis found that since January 2026, the best open-weight models have trailed frontier closed models by an average of four months, or 8 ECI points on the Epoch Capabilities Index, a gap comparable to the difference between GPT-5 and GPT-5.5. Within that context, Moonshot AI released Kimi K3, described as the first open-weight 2.8-trillion-parameter frontier model, publishing weights to Hugging Face in a repository totaling about 1.56 terabytes under a custom license. On consumer hardware, Simon Willison reported that Qwen3.6-35B running on a laptop produced better SVG art than Claude Opus 4.7, and that Qwen 3.8 27B was the first local model he felt was nearly competitive with frontier systems, writing, 'I thought I'd have to wait five years and spend ten thousand dollars on hardware to get results even half as good as this one.' PrismML compressed a 27B-parameter model to 5.9 GB while retaining 98.2% of benchmark performance, and Stanford research cited by Milk Road AI found inference costs for GPT-3.5-level models fell more than 280-fold in roughly 18 months. Anthropic CEO Dario Amodei publicly stated his company is not pushing for a prohibition on open-weight models.