Models / Best of /open-source
Updated August 2026 · ranked from live data

Best open-weight models

Fully open weights (no restricted custom licenses), ranked by measured intelligence rating, then recency.Right now the top pick is GLM 5.2 (4/5 intelligence (Arena Text (overall), 18 of 143)).Every row links to full specs, hardware verdicts and provider pricing.

1GLM 5.2Z.AI4/5 intelligence (Arena Text (overall), 18 of 143)753B$0.63 / $1.98per 1M · via OpenRouter2MiMo-V2.5-ProXiaomi4/5 intelligence (Arena Text (overall), 23 of 143)1T$0.43 / $0.87per 1M · via OpenRouter3GLM 5.1Z.AI4/5 intelligence (Arena Text (overall), 21 of 143)754B$0.97 / $3.04per 1M · via OpenRouter4Hy3Tencent3.5/5 intelligence (Arena Text (overall), 31 of 143)299B$0.13 / $0.53per 1M · via OpenRouter5DeepSeek V4 ProDeepSeek3.5/5 intelligence (Arena Text (overall), 28 of 143)1.6T / 37B$0.43 / $0.87per 1M · via DeepSeek6Gemma 4 31BGoogle3.5/5 intelligence (Arena Text (overall), 33 of 143)32.7B$0.10 / $0.34per 1M · via OpenRouter7GLM 5Z.AI3.5/5 intelligence (Arena Text (overall), 29 of 143)754B$0.95 / $2.55per 1M · via OpenRouter8InklingThinking Machines3/5 intelligence (Arena Text (overall), 45 of 143)952B$1.00 / $4.05per 1M · via OpenRouter9MiMo-V2.5Xiaomi3/5 intelligence (Arena Text (overall), 50 of 143)311B$0.14 / $0.28per 1M · via OpenRouter10DeepSeek V4 FlashDeepSeek3/5 intelligence (Arena Text (overall), 48 of 143)291B / 37B$0.14 / $0.28per 1M · via Novita AI
More listsBest models for codingCheapest hosted modelsBest models that fit in 12 GB VRAMBest models for 24–32 GB VRAMLongest context windowsFastest hosted models