GLM-5.3

Data updated 12 Sept 2026

18 published benchmark measures · 6 benchmark families contribute across 3 task areas. 3 capability estimates available in the full profile. See all results ↓ · Compare published benchmarks →

Model evidence summary

Choose a configuration

A model-wide score would mix effort settings. This page summarizes published evidence; only identified configurations receive leaderboard ranks.

Compare measured configurations →

The capability chart summarizes evidence across reported settings, not a runnable configuration or a leaderboard rank.

Performance profile

Capabilities

Adjusted comparison score · 0–100

Filled points are supported; hollow points are preliminary. Lines stop at unknown capabilities. Exact values and sources follow below.

Agentic2 families · 0 with independent evidence · Supported68.6

2 core families; 6 direct opponents across 6 labs.

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: 65.2–69.9; 2/11 scenarios unsupported. Without one publisher: 65.7–71.3; 1/15 unsupported. Smoothing check: 64.1–71.0; 0/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: family and publisher holdouts improved on neutral predictions within evaluated coverage; source sensitivity still requires inspection. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

AutomationBench: 100.0 observed win share
Z.ai · Source 1
Toolathlon: 33.3 observed win share
Z.ai · Source 1

Hard reasoning2 families · 1 with independent evidence · Supported64.2

2 core families; 25 direct opponents across 12 labs.

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: 60.0–65.3; 2/7 scenarios unsupported. Without one publisher: 58.0–66.7; 2/16 unsupported. Smoothing check: 60.9–65.8; 0/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: family and publisher holdouts improved on neutral predictions within evaluated coverage. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

GPQA: 31.8 observed win share
artificialanalysis.ai · Source 1
Humanity’s Last Exam: 83.3 observed win share
Z.ai · Source 1

Coding2 families · 1 with independent evidence · Supported69.6

2 core families; 27 direct opponents across 9 labs.

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: 65.5–70.3; 2/7 scenarios unsupported. Without one publisher: 66.9–71.3; 1/20 unsupported. Smoothing check: 65.4–71.5; 0/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: family and publisher holdouts improved on neutral predictions within evaluated coverage. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

DeepSWE: 72.2 observed win share
Z.ai, deepswe.datacurve.ai · Source 1 Source 2
Terminal-Bench: 70.8 observed win share
Z.ai · Source 1 Source 2

Human prefNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 1/1 scenarios unsupported. Without one publisher: No supported estimate; 1/1 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: cross-family and cross-publisher predictions could not be evaluated from the available comparison network. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

KnowledgeNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 4/4 scenarios unsupported. Without one publisher: No supported estimate; 5/5 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: almost no publisher-held-out comparisons remained estimable. Cross-source validity is unresolved. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

MultimodalNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 8/8 scenarios unsupported. Without one publisher: No supported estimate; 8/8 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: limited publisher-held-out coverage. Cross-source validity is unresolved. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

Long contextNo comparable evidenceUnknown

0 core families; 0 direct opponents across 0 labs. No admitted core comparisons

Estimated against the same complete panel: gpt-5.6-sol claude-opus-4-8 gemini-3.1-pro . Indirect comparisons assume performance can be summarized by one strength within this capability.

Refit without one family: No supported estimate; 5/5 scenarios unsupported. Without one publisher: No supported estimate; 5/5 unsupported. Smoothing check: No supported estimate; 3/3 unsupported. These are sensitivity checks, not confidence intervals.

Audit: family and publisher holdouts underperformed neutral predictions. Treat this as an exploratory estimate. Validation audit.

Matched family win shares below describe source evidence; they are not averaged to produce the adjusted score.

Scores estimate outcomes against a shared reference panel; they are not accuracy percentages. Sparse or disconnected evidence cannot qualify an overall profile. Open a capability to inspect its evidence.

Compare 3 effort levels across 36 benchmark/harness combinations →

Reported effort · Max + unspecified

Settings reported in this model's published benchmark results, including results outside the aggregate. Effort names are provider-specific. These are not API defaults or equal compute budgets.

Max: 9 observations
Not specified: 13 observations

Mixed settings means multiple settings occur in the evidence. Best across efforts means the source selected its best reported result across settings; it does not mean Max. Unspecified settings stay unknown. This model-summary chart combines reported settings. The leaderboard keeps identified configurations separate and excludes unknown effort.

Inspect each result and its source ↓ · Download effort evidence

Compare capability profiles →

Score contributions and missing evidence

6 contributing families across 3 capabilities. Fixed reference panels do not change when the catalog expands.

Capability is fitted jointly across families. Capability estimates below describe different task areas; their weighted sum is not the Capability score.

Results without reviewed compatibility or a reference match remain in the raw evidence below. Coverage counts only contributing results.

Model information & shareable badge

Lab: Z.ai
Catalog status: active
Availability: Documented provider API; downloadable official weights
Family: GLM
Released: —
Context: 1,000,000 tokens
License: —
Model card: https://docs.z.ai/guides/overview/overview
Default Capability family coverage: /badge/glm-5.3.svg

Benchmark scores & sources

Original results, evaluation harnesses, and evidence behind this model.

Benchmark	Bucket	Score	Harness	Evidence	Source-recorded date
Agents' Last Exam (ALE-CLI) · source release snapshot; version not specified	Supporting evidence	28.5% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: Agents' Last Exam (ALE-CLI) · reviewed 2026-09-12	Published configuration Effort: Max	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
AutomationBench (v1.0.6) · v1.0.6	Agentic	48.2% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: AutomationBench (v1.0.6) · reviewed 2026-09-12	Published configuration Effort: Not specified contributes to capability	lab self-report	Reviewed 2026-09-12
CyberGym · source release snapshot; version not specified	Supporting evidence	84.5% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: CyberGym · reviewed 2026-09-12	Published configuration Effort: Max	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
DeepSWE (v1.1) · v1.1	Coding	66.9% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: DeepSWE (v1.1) · reviewed 2026-09-12	Published configuration Effort: Not specified contributes to capability	lab self-report	Reviewed 2026-09-12
DeepSWE v1.1	Coding	69%	DeepSWE v1.1 reported Effort: Not specified contributes to capability	official board	2026-09-03
ExploitBench · source release snapshot; version not specified	Supporting evidence	54.4% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: ExploitBench · reviewed 2026-09-12	Published configuration Effort: Max	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
ExploitGym (2h / 6h) · source release snapshot; version not specified	Supporting evidence	105 count Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. 2h Single-run869tasks; API time normalized by model TPS plus overhead; reported number solved; domain whitelist. First-party reported result. zai-org/GLM-5.3 · Performance table: ExploitGym (2h / 6h) · reviewed 2026-09-12	Published configuration Effort: Max	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
ExploitGym (2h / 6h) · source release snapshot; version not specified	Supporting evidence	130 count Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. 6h Single-run869tasks; API time normalized by model TPS plus overhead; reported number solved; domain whitelist. First-party reported result. zai-org/GLM-5.3 · Performance table: ExploitGym (2h / 6h) · reviewed 2026-09-12	Published configuration Effort: Max	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
FrontierSWE · source release snapshot; version not specified	Supporting evidence	78.1% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. Externally evaluated result, attributed in the source footnote. Comparison limit: The GLM card attributes this evaluation to Proximal; it is not a new Z.ai comparison. zai-org/GLM-5.3 · Performance table: FrontierSWE · reviewed 2026-09-12	Published configuration Effort: Max	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
GDPval-AA v2 · source release snapshot; version not specified	Agentic	1769 Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. Externally evaluated result, attributed in the source footnote. Comparison limit: The GLM card attributes this evaluation to Artificial Analysis; it is not a new Z.ai comparison. zai-org/GLM-5.3 · Performance table: GDPval-AA v2 · reviewed 2026-09-12	Published configuration Effort: Not specified	lab self-report The GLM card attributes this evaluation to Artificial Analysis; it is not a new Z.ai comparison.	Reviewed 2026-09-12
GPQA Diamond	Hard reasoning	91.717%	GPQA Diamond reported Effort: Not specified contributes to capability	independent repro	2026-09-11
HLE w/ Tools · source release snapshot; version not specified	Hard reasoning	62.5% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: HLE w/ Tools · reviewed 2026-09-12	Published configuration Effort: Not specified contributes to capability	lab self-report	Reviewed 2026-09-12
NL2Repo · source release snapshot; version not specified	Supporting evidence	58% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: NL2Repo · reviewed 2026-09-12	Published configuration Effort: Not specified	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
PostTrainBench · source release snapshot; version not specified	Supporting evidence	39.8% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. Comparison limit: The source changes anti-cheat checks and substitutes zero-shot baseline scores for failed runs; the modified protocol requires separate review. zai-org/GLM-5.3 · Performance table: PostTrainBench · reviewed 2026-09-12	Published configuration Effort: Max	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
ProgramBench (Almost Solved) · source release snapshot; version not specified	Supporting evidence	19% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: ProgramBench (Almost Solved) · reviewed 2026-09-12	Published configuration Effort: Not specified	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
SWE-Marathon (v1.1) · v1.1	Supporting evidence	42.5% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. Comparison limit: The source replaces affected anti-cheat checks with LLM inspection and changes task images; the modified protocol requires separate review. zai-org/GLM-5.3 · Performance table: SWE-Marathon (v1.1) · reviewed 2026-09-12	Published configuration Effort: Max	lab self-report Supporting evidence outside the reviewed capability core	Reviewed 2026-09-12
Terminal Bench 2.1 · 2.1	Coding	88.2% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: Terminal Bench 2.1 · reviewed 2026-09-12	Published configuration Effort: Not specified contributes to capability	lab self-report	Reviewed 2026-09-12
Terminal Bench 3.0 · 3.0	Coding	28.3% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: Terminal Bench 3.0 · reviewed 2026-09-12	Published configuration Effort: Max contributes to capability	lab self-report	Reviewed 2026-09-12
Toolathlon Verified · source release snapshot; version not specified	Agentic	73% Reported settings & source GLM-5.3 release table; benchmark-specific harness and reasoning configurations in Evaluation Details. First-party reported result. zai-org/GLM-5.3 · Performance table: Toolathlon Verified · reviewed 2026-09-12	Published configuration Effort: Not specified contributes to capability	lab self-report	Reviewed 2026-09-12
VulcanBench v3	Supporting evidence	73.9%	VulcanBench v3 bare-bones API · Report 18 · high Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-08-24
VulcanBench v3	Supporting evidence	78.3%	VulcanBench v3 bare-bones API · Report 18 · low Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-08-24
VulcanBench v3	Supporting evidence	65.2%	VulcanBench v3 bare-bones API · Report 18 · max Effort: Not specified	official board Supporting evidence outside the reviewed capability core	2026-08-24
ARC-AGI-2	Hard reasoning	—	—	—	—
GDPval-AA	Agentic	—	—	—	—
Humanity's Last Exam	Hard reasoning	—	—	—	—
LiveCodeBench	Coding	—	—	—	—
LMArena Text Arena	Human pref	—	—	—	—
MMLU-Pro	Knowledge	—	—	—	—
OSWorld-Verified	Agentic	—	—	—	—
SWE-bench Pro	Agentic	—	—	—	—
SWE-bench Verified	Agentic	—	—	—	—
Terminal-Bench 2.1	Agentic	—	—	—	—