THE FORGE RANKINGS·2026-08-30

o3 (2025-04-16)

OpenAI
29
RANK of 37 ranked
FERROX INDEX42.2 weighted mean of category positions
GRADED back of the field
EVIDENCE11 / 32 benchmarks measured
QUALITYFULL every category rests on independent benchmarks

Not separable from MiniMax-M3 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

23of 37 measured
36.6
13.1median 38.963.0

Coding

8of 37 measured
55.3
22.4median 50.070.7

Reasoning

31of 37 measured
38.3
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.

ALE-Bench

25/35
934 best 2177 · GPT-5.6 Sol
ALE-Bench (Sakana AI with AtCoder) CC-BY-4.0 high

APEX Agents

24/28
17.2% best 45.0% · Claude Fable 5
APEX CC-BY-4.0 high

SWE-bench Verified (Epoch's own run)

19/21
62.3% best 83.5% · Claude Opus 4.7
Epoch AI CC-BY-4.0 medium

LMArena Text (style-controlled)

16/20
1431 best 1507 · Claude Fable 5
LMArena CC-BY-4.0

Not measured

21/32
  • LMArena Agent
  • DeepSWE
  • FrontierCode
  • Terminal-Bench 2
  • LMArena WebDev
  • Terminal-Bench 4.0
  • 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
APEX Agents agents 17.2% 12.1 to 22.3 APEX CC-BY-4.0 / high source
METR time horizons agents 1.5h 47.4 to 63.2 METR - Measuring AI Ability to Complete Long Tasks CC-BY-4.0 / medium source
Aider Polyglot coding 76.9% not published Aider LLM Leaderboards CC-BY-4.0 diff / medium source
ALE-Bench coding 934 not published ALE-Bench (Sakana AI with AtCoder) CC-BY-4.0 / high source
SWE-bench Verified (Epoch's own run) coding 62.3% 58.0 to 66.7 Epoch AI CC-BY-4.0 / medium source
LMArena Text (style-controlled) preference 1431 75.4 to 76.4 LMArena CC-BY-4.0 source
ARC-AGI-2 reasoning 3.0% not published ARC Prize CC-BY-4.0 / medium source
FrontierMath reasoning 16.9% 12.6 to 21.2 Epoch AI CC-BY-4.0 / medium source
GPQA Diamond (Epoch's own run) reasoning 80.8% 75.3 to 86.3 Epoch AI CC-BY-4.0 / medium source
Humanity's Last Exam reasoning 19.2% 16.2 to 22.2 Humanity’s Last Exam (CAIS / Scale AI) CC-BY-4.0 / medium source
OTIS Mock AIME 2024-2025 reasoning 83.9% 75.3 to 92.5 Epoch AI CC-BY-4.0 / high source

Configurations rolled up: 7. Rule: median observed configuration per benchmark (lower median, always a real measurement). Harnesses seen: diff.

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.