Skip to main content
EvidenceAI ModelsVisualization detail

Terminal-Bench / standard coding benchmarks.

Standard coding benchmarks

Takeaway: A few points behind Opus/GPT, well ahead of Gemini.

Standard coding

Terminal-Bench 2.1, SWE-bench Pro, NL2Repo, DeepSWE, ProgramBench.

Takeaway: A few points behind Opus/GPT, well ahead of Gemini.

Standard coding

A few points behind Opus/GPT, well ahead of Gemini.

Rendering the interactive chart…

Expand data table
Download CSV
Standard coding benchmarks: data table
DimensionSeriesValueValue basisSource
Terminal-Bench 2.1Glm5163.5ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
Terminal-Bench 2.1Glm5281ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
Terminal-Bench 2.1Gpt5584ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
Terminal-Bench 2.1Gemini74ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
Terminal-Bench 2.1Opus4885ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
SWE-bench ProGlm5158.4ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
SWE-bench ProGlm5262.1ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
SWE-bench ProGpt5558.6ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
SWE-bench ProGemini54.2ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
SWE-bench ProOpus4869.2ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
NL2RepoGlm5142.7ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
NL2RepoGlm5248.9ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
NL2RepoGpt5550.7ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
NL2RepoGemini33.4ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
NL2RepoOpus4869.7ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
DeepSWEGlm5118ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
DeepSWEGlm5246.2ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
DeepSWEGpt5570ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
DeepSWEGemini10ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
DeepSWEOpus4858ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
ProgramBenchGlm5150.9ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
ProgramBenchGlm5263.7ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
ProgramBenchGpt5570.8ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
ProgramBenchGemini39.5ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
ProgramBenchOpus4871.9ActualZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)

Standard coding benchmarks: full data table

SOURCEZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)
VALUE BASISActual
LAST VERIFIEDSep 10, 2026
How this is calculated

Data table

Download CSV
Standard coding benchmarks: coding_std data table (GLM-5.2 Benchmarks)
Glm51Glm52Gpt55GeminiOpus48SeriesBenchmarkSourceValue basis
63.581847485coding_stdTerminal-Bench 2.1Z.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)Actual
58.462.158.654.269.2coding_stdSWE-bench ProZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)Actual
42.748.950.733.469.7coding_stdNL2RepoZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)Actual
1846.2701058coding_stdDeepSWEZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)Actual
50.963.770.839.571.9coding_stdProgramBenchZ.ai: GLM-5.2: Built for Long-Horizon Tasks (2026-06-16)Actual

Method

How this evidence was built

What it measures

Source-backed values for all six charts come from the Z.ai glm-5.2 launch post (2026-06-16): long-horizon coding (FrontierSWE, PostTrainBench, swe-Marathon), standard coding (Terminal-Bench 2.1, swe-bench Pro, NL2Repo, DeepSWE, ProgramBench), reasoning (hle, CritPt, aime, hmmt, gpqa-Diamond, IMOAnswerBench), agentic (mcp-Atlas, Tool-Decathlon), the glm-5.2 vs glm-5.1 generational leap, and the mtp acceptance-length ablation.

Key context

  • Every numeric point carries a source citation and a basis label naming the table row.
  • Scores are reproduced verbatim from the Full Benchmark Table; cells the vendor published as "-" (not reported) are omitted, never estimated.

Key numbers

  • FrontierSWE: glm-5.2 74.4 vs Opus 4.8 75.1 vs gpt-5.5 72.6 vs Gemini 39.6 vs glm-5.1 30.5 (zai-glm52).
  • Terminal-Bench 2.1: glm-5.2 81 vs Opus 4.8 85 vs gpt-5.5 84 vs Gemini 74 (zai-glm52).
  • Generational leap: swe-Marathon 13.0 (5.2) vs 1.0 (5.1) (zai-glm52).
  • Mtp ablation quantifies the multi-token-prediction contribution to accepted length (zai-glm52).

Caveats

Caveat: these are the model developer’s self-reported figures under their stated harnesses and prompts (see the post’s footnotes); some competitor hle/CritPt cells are full-set scores. They are comparison-as-published, not an independent eval. Architecture context (not charted): 1M-token context (up from 200K), IndexShare cuts per-token indexer FLOPs ~2.9× at 1M, mit-licensed open weights.

Caveat: Some competitor scores are marked full-set vs public-set, not always like-for-like.

Do not use this to

  • Independent verification: every figure is vendor self-reported (zai-glm52).
  • Real-world reliability, latency or cost (benchmark scores only).

Scope

June 2026

Sources

Last verified 10 Sept 2026

How relevant was this information?