GrindBenchThe daily grind, benchmarked.

Rankings

28 AI models, ranked by how much real work they got right. Pick a benchmark, then a kind of work. Switch to the chart to see what each one costs.

Results as of 2026-11-03. "Right" means no deal-breaker mistakes and at least 80% of the checklist.

Each model's share of work right, averaged over PubSec and Banking with each counting equally, and its average cost per task. Only models that have run every benchmark are ranked.

  1. Group 1 · too close to call within a group1BClaude Sonnet 5.5Anthropic75%69–81%8.7¢ a task
  2. 2BGLM-5.3Z.ai (Zhipu) · open weights69%63–76%$0.16 a task
  3. 3BMuse Spark 1.3Meta64%57–71%4.2¢ a task
  4. 4BDeepSeek V4.1 FlashDeepSeek · open weights63%56–70%1¢ a task
  5. Group 25BGPT-6.1 SolOpenAI61%54–67%3.4¢ a task
  6. 6CMercury 3Inception56%49–63%0.94¢ a task
  7. 7CGLM-5.3 FlashZ.ai (Zhipu)54%47–61%0.99¢ a task
  8. 8CQwen3.8 FlashAlibaba (Qwen) · open weights51%43–58%1.4¢ a task
  9. 9CGemini 3.8 FlashGoogle50%43–57%6.5¢ a task
  10. Group 310CMiniMax M3MiniMax · open weights45%38–51%2.4¢ a task
  11. 11DNemotron 3 UltraNVIDIA · open weights38%32–44%4¢ a task
  12. 12DGPT-6 LunaOpenAI36%31–42%0.24¢ a task
  13. Group 413DNemotron 3 SuperNVIDIA · open weights22%18–26%0.69¢ a task
  14. 14FQwen3.6 35B A3BAlibaba (Qwen) · open weights18%14–22%1.2¢ a task
  15. Group 515FNemotron 3.5 LightningNVIDIA · open weights13%9–16%0.53¢ a task
  16. 16FGemma 4 31BGoogle · open weights10%7–13%0.11¢ a task
  17. 17FGemma 4 26B A4BGoogle · open weights9%6–11%0.082¢ a task
  18. 18FMistral Small 4Mistral AI · open weights7%5–10%0.19¢ a task
10 more models that haven't run every benchmark

Unranked here: an average over different benchmarks wouldn't be a fair comparison. Pick a benchmark above to see each one ranked.

Next

Find your own work

Pick a job and see what AI can do for it today, how often it gets it right, and what it costs.