skynet · coding
SKYNET

zai-coding/glm-5.2

ZAI flagship coding

⚖ COMPARE MODELS
VALS INDEX
30.5% +/- 25.4
Accuracy
VALS INDEX
25.2s
Latency (median)
$VALS INDEX
$0.2921
Cost / Test (1P-est)
SCORE STATUSscored
TEST SETS13
RECORDED TRIALS708
CUSTOM PASS50%
DEEP VALIDATION · 2026-08-20

Dual-judge family campaign

FULL REPORT →
CONSENSUS MEAN3.878 / 5
BOTH-JUDGE PASS75.6%
CROSS-JUDGED45 / 45
MEDIAN WALL18.87s
GPQA_DIAMONDserialized verification sample
52.0%
23/25 responses
IFEVALserialized verification sample
76.0%
25/25 responses
MMLU_PROserialized verification sample
84.0%
25/25 responses

Industry benchmarks

Accuracyattempts · responded
GPQA_DIAMOND GPQA Diamond
22.3%
202 attempts · 126 responded
HUMANEVAL_PLUS HumanEval+
68.0%
50 attempts · 40 responded
IFEVAL IFEval
12.4%
140 attempts · 26 responded
MMLU_PRO MMLU-Pro
19.2%
260 attempts · 169 responded
Industry benchmark Bars = accuracy %, black cap = that score

Custom 9-task suite

Mean /5n=trials
CUSTOM code_gen_long
5.00
8 trials
CUSTOM json_strict
5.00
3 trials
CUSTOM reasoning_multistep
5.00
3 trials
CUSTOM summarize
4.62
8 trials
CUSTOM agentic_prompt
3.50
8 trials
CUSTOM code_debug -
2.38
8 trials
CUSTOM refactor_existing_code
1.80
5 trials
CUSTOM creative_write -
0.62
8 trials
CUSTOM agentic_tool_use -
0.00
5 trials
Custom suite · scored 0–5 by consensus (m3 + Fable-5) ★ = task scored a 5 at least once · sorted by mean score

Cross-judge validation

m3 vs Fable-5 Δ

No judge bias detected on this model — m3 and Fable-5 agree across all tasks.

Per-task distribution

Score

The accumulated custom suite below uses minimax-m3 + claude-fable-5 consensus. The dedicated 2026-08-20 campaign above is supplemental and uses minimax-m3 + a blind grok-4.6 re-judge on every successful response. Industry suites use deterministic family-specific graders. See methodology and the campaign report.