THE FORGE RANKINGS·2026-08-30

GLM-5.3

Z.ai (Zhipu AI)
8
RANK of 37 ranked
FERROX INDEX84.9 weighted mean of category positions
GRADEA front rank
EVIDENCE6 / 32 benchmarks measured
QUALITYFULL every category rests on independent benchmarks

Not separable from Claude Opus 4.7 at the measured 1.78 point band. The order is printed; the gap is not claimed.

No measurement in Preference. That weight was redistributed, which is the same as assuming this model would have scored its own average there. It is an assumption, not a measurement.

Compared to what

Each rail is one classification. Every ranked model measured on it is a tick; this model is the orange marker. A hatched rail means this model has no measurement in that classification and its weight was redistributed.

Agents

25of 37 measured
35.8
13.1median 38.963.0

Coding

12of 37 measured
53.3
22.4median 50.070.7

Reasoning

3of 37 measured
91.0
18.1median 62.893.0

Preference

no qualifying measurement
53.1median 71.882.5

Every measurement, and the field behind it

One card per benchmark this model has been measured on. The rail shows where it sits against every other ranked model measured on the same benchmark. Where the evaluator keys on a harness, the harness is named, because the same model scores differently under a different one.

GPQA Diamond (Epoch's own run)

7/36
90.9% best 94.1% · Gemini 3.1 Pro Pr…
Epoch AI CC-BY-4.0 max

OTIS Mock AIME 2024-2025

17/36
91.1% best 99.7% · Claude Fable 5
Epoch AI CC-BY-4.0 max

ALE-Bench

15/35
1317 best 2177 · GPT-5.6 Sol
ALE-Bench (Sakana AI with AtCoder) CC-BY-4.0 high

LMArena Agent

12/21
0.039 best 0.127 · Claude Opus 5
LMArena CC-BY-4.0

DeepSWE

3/17
69.0% best 72.8% · Claude Opus 5
deepswe.datacurve.ai CC-BY-4.0 mini-swe-agent max

Terminal-Bench 4.0

3/10
41.8% best 51.8% · Claude Opus 5
Terminal-Bench Apache-2.0 claude-code max

Not measured

26/32
  • ARC-AGI-2
  • APEX Agents
  • FrontierMath
  • SWE-bench Verified (Epoch's own run)
  • LMArena Text (style-controlled)
  • FrontierCode
  • Humanity's Last Exam
  • Terminal-Bench 2
  • LMArena WebDev
  • METR time horizons
  • Aider Polyglot
  • OSWorld
  • Cybench
  • BrowseComp
  • IMOAnswerBench
  • SWE-bench Multilingual
  • AIME
  • CyberGym
  • HMMT
  • MCP-Atlas
  • SWE-bench Pro
  • Tau2-Bench Airline
  • Tau2-Bench Banking
  • Tau2-Bench Retail
  • Tau2-Bench Telecom
  • Tool-Decathlon
A missing benchmark is not a zero and is never scored as one.

Check it yourself

Every number on this page, with who measured it, the interval they published, the harness it was run under, and where to go and read it. Nothing here was measured by Ferrox Labs.

BenchmarkClassPublishedInterval (normalised) Measured byLicenceHarness / effort
LMArena Agent agents 0.039 26.6 to 32.8 LMArena CC-BY-4.0 source
Terminal-Bench 4.0 agents 41.8% 38.6 to 45.0 Terminal-Bench Apache-2.0 claude-code / max source
ALE-Bench coding 1317 not published ALE-Bench (Sakana AI with AtCoder) CC-BY-4.0 / high source
DeepSWE coding 69.0% 65.9 to 72.0 deepswe.datacurve.ai CC-BY-4.0 mini-swe-agent / max source
GPQA Diamond (Epoch's own run) reasoning 90.9% 87.8 to 94.0 Epoch AI CC-BY-4.0 / max source
OTIS Mock AIME 2024-2025 reasoning 91.1% 84.3 to 97.9 Epoch AI CC-BY-4.0 / max source

Configurations rolled up: 4. Rule: median observed configuration per benchmark (lower median, always a real measurement). Harnesses seen: claude-code, mini-swe-agent.

Snapshot 2026-08-30-62e4e85f043b, manifest dc46131315046f1e. Grades are positional: position in the measured field, as a percentile of rank among ranked entries, n=37.