ModelCards
← All benchmarks

PostTrainBench

We haven't written a description for this benchmark yet; see the card notes below for how each maker ran it.

Your workloadRAG over docs · 12K in · 600 out · 50K requests/mo · 30% cached
  1. 1Gemini 4 Argon · Googlebest value
    45.3%
  2. 2GLM-5.3 · Z.aibest value
    39.8%

“Best value”: no other model here reports a higher score at a lower cost for your workload. Scores are self-reported by each maker with different settings, so treat close calls as ties.

How each maker reported it

  • Gemini 4 Argon Evaluation

    Results as of October 2026, as published in Google's model card; methodology at deepmind.google/models/evals-methodology/gemini-4-argon. Comparison scores are Google's figures. CWE-bench v1 is a tie with GPT-6 Astra.

  • GLM-5.3 model card, Benchmark table

    Z.ai's own table (GLM-5.3 at max effort). Most agentic runs use the Claude Code harness; FrontierSWE was run by Proximal and GDPval-AA by Artificial Analysis. ExploitGym is tasks solved within a 6-hour normalized budget. Qwen3.8-Max and DeepSeek-V4 Pro columns omitted here.

Only each model's own maker's number is shown; scores a maker reports for rival models are left out.