Zhipu AIChina· released Jun 2026· open weights

GLM 5.2

Name: GLM 5.2
Author: Zhipu AI

Ranked #24, 36.7% behind the frontier on average.

Zhipu AI, China. One of 7 models from this lab on the Index.

Third-party publishedMedium confidence (7+ comparable tests, 3+ Index categories)

Snapshot September 8, 2026

OpenCharts Index

63.3

Ranked #24 (21 to 28) of 36 · 36.7% behind the frontier on average · 19 comparable tests · 3 counting categories

Drop any single test and the Index lands between 59.6 and 66.9, anywhere from #21 to #28. Neighbours inside that range are ties.

4 of 8 hardest tests · share of the frontier

Distance to today's frontier on the hardest tests. Not a measure of intelligence.

What it's known for

GLM 5.2, in 19 results.

Strongest in coding (69.6) and reasoning (66.6). It trails the frontier most in math, 46.3% behind on average. No comparable results yet in multimodal and long context. Between 35% and 40% behind the best published results, on average. Drop any single test and it would sit anywhere from #21 to #28.

Strongest category

Coding

Writing, fixing and shipping real code, judged by tests or users.

69.6

SWE-bench Verified78.7% · 94% of best
SciCode50.5% · 81% of best
WeirdML70.1% · 76% of best
CursorBench55.0% · 75% of best
Code Arena (WebDev)1,587 · 46% win rate vs leader
FrontierCode24.5% · 46% of best

Mean of 6 of the category's 7 comparable tests. Counts toward the Index.

Best single result

GPQA Diamond

Answer a PhD-level multiple-choice question in physics, chemistry or biology that a web search will not settle.

91.9%

94.5% of the best (95.8%), above chance · #16 of 78.

Expert-level science questions are the closest thing to asking a specialist colleague. A model that gets them right can check a technical claim instead of echoing it.

Where it trails

ARC-AGI-2

Infer the hidden rule from a few input and output grid pairs, then apply it to a new grid.

22.8%

76.0% behind the best published result (95.0%) · #34 of 51.

Novel visual puzzles resist memorization, so they measure fluid intelligence rather than recall.

Not measured yet

No comparable result yet in these 2 categories. A blank is a blank, never a zero, and it does not lower the Index.

Multimodal
Long context

Covered, but not averaged into the Index:

Knowledgesingle-test category, shown beside the Index · 45.2
Agentic2 of 8 tests · needs 4 · mean 82.8
Human preferencesingle-test category, shown beside the Index · 89.8

Every score, with its source

See it at work

What GLM 5.2 was asked to do, and how it did.

Every test it has taken, by category, with the kind of task it faced, who produced the number and where the result landed against the best published one. The examples are original and illustrative, never items from the datasets themselves.

Human preference · mean 89.8 across 1 comparable test · a single-test category, shown beside the Index and never averaged into it

Text ArenaLeaderboard

Answer the same prompt as a rival model; real users vote blind on which answer they prefer.

1,472

45% expected win rate against the board leader (1,507)
#29 of 92

max · board 2026-09-02

Source: Arena

Every result is read against the best published one on its test: percent scores as a share of the best above chance, Arena ratings as the expected win rate against the board leader, open-ended values on a log scale. Sitting a harder exam never lowers a model. Every number links to the publisher that produced it.

Head to head

GLM 5.2 against whoever you pick.

Choose a rival. Every counted test both have taken appears side by side as a share of the best published result on that test, with the category means above. A win is a higher share; a gap under 2 points is a tie, because one item on a 45-problem exam is 2.2 points and publishers' own standard errors are of that order.

GLM 5.2

#24·Index 63.3C

0 : 14

wins · 0 ties · 14 shared

GPT-6 Astra

#1·Index 99.0A+

GPT-6 Astra comes out ahead on 14 of the 14 tests both have taken.

Category means

Reasoning
66.6
99.6
Knowledge
45.2
100.0
Coding
69.6
97.7
Math
53.7
99.8
Agentic
82.8
98.5
Human preference
89.8
—

Shared tests · biggest gaps first

ARC-AGI-2Reasoning
22.8%
95.0%
−76.0
FrontierMath Tier 4Math
29.3%
97.6%
−70.0
ProofBenchMath
35.0%
99.0%
−64.0
SimpleQA VerifiedKnowledge
34.2%
75.6%
−54.8
Code Arena (WebDev)Coding
1,587
1,797
−54.1
FrontierCodeCoding
24.5%
53.3%
−53.8
FrontierMath (Tiers 1–3)Math
59.2%
93.7%
−36.8
CritPtReasoning
20.9%
31.7%
−33.6
WeirdMLCoding
70.1%
92.9%
−24.5
APEX-AgentsAgentic
35.6%
46.7%
−23.4
ARC-AGI-1Reasoning
77.0%
98.5%
−21.8
Mock AIME 2024–2025Math
86.4%
100.0%
−13.6
SciCodeCoding
50.5%
56.5%
−9.7
GPQA DiamondReasoning
91.9%
95.8%
−5.5

Bars are distances to the best published result on each test (percent scores above chance, Arena ratings as win rate against the leader, open-ended values on a log scale). Raw values as the publishers report them.

Open GPT-6 Astra

Scores

Every test, every source

The best published run per benchmark, the raw value as the publisher reports it, who produced the number, how close it comes to the best published result on that test, and where the model sits among every model scored on it. Percent scores are read above chance, Arena ratings as a win rate against the board leader, open-ended values on a log scale. A test with too few models to compare against is shown but not counted.

Category means

Reasoning66.6 · 5/6
Knowledgebeside45.2 · 1/1
Coding69.6 · 6/7
Math53.7 · 4/5
Agentic82.8 · 2/8
Multimodal—
Human preferencebeside89.8 · 1/1
Long context—

Mean · tests taken of the category's comparable tests. A faded category either rests on fewer than half of them or is a single-test category (marked “beside”), which is shown next to the Index and never averaged into it.

GPQA DiamondEpoch-runReasoning

91.9%

94.5% of the best (95.8%), above chance
81th percentile of 78 · n = 198

max

GLM 5.2

GLM 5.2, in 19 results.

What GLM 5.2 was asked to do, and how it did.

GLM 5.2 against whoever you pick.

Every test, every source

Compare with its neighbours.

The best models, ranked here, working inside Theo.