The open-weight model Qwen3.8 Max has achieved a score of 45.4, while MiMo-V2.6-Pro reached 46.3. Both models are being compared against the proprietary GPT-6 Astra, which scored 52.7. Qwen3.8 Max is also noted to be significantly cheaper, costing 8 times less per 1 million output tokens compared to GPT-6 Astra, and MiMo-V2.6-Pro is 57 times cheaper. AI
IMPACT Open-weight models are demonstrating competitive performance and cost-effectiveness against proprietary systems.
RANK_REASON Comparison of open-weight models against a proprietary model on benchmarks.
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →