THE FORGE RANKINGS·2026-08-30

Kimi K2.5

Moonshot
28
RANK of 37 ranked
FERROX INDEX43.4 weighted mean of category positions
GRADED back of the field
EVIDENCE6 / 32 benchmarks measured
QUALITYFULL every category rests on independent benchmarks

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

19of 37 measured
38.9
13.1median 38.963.0

Coding

21of 37 measured
48.6
22.4median 50.070.7

Reasoning

37of 37 measured
18.1
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

28/35
822 best 2177 · GPT-5.6 Sol
ALE-Bench (Sakana AI with AtCoder) CC-BY-4.0

ARC-AGI-2

20/29
11.8% best 88.3% · Claude Opus 5
ARC Prize CC-BY-4.0

APEX Agents

25/28
14.4% best 45.0% · Claude Fable 5
APEX CC-BY-4.0

SWE-bench Verified (Epoch's own run)

11/21
73.8% best 83.5% · Claude Opus 4.7
Epoch AI CC-BY-4.0

Humanity's Last Exam

8/17
24.4% best 46.4% · Gemini 3.1 Pro Pr…
Humanity’s Last Exam (CAIS / Scale AI) CC-BY-4.0

OSWorld

4/6
63.3% best 75.2% · MiniMax-M3
OSWorld (XLANG Lab) none stated General model 100 steps

Not measured

26/32
  • GPQA Diamond (Epoch's own run)
  • OTIS Mock AIME 2024-2025
  • LMArena Agent
  • FrontierMath
  • LMArena Text (style-controlled)
  • DeepSWE
  • FrontierCode
  • Terminal-Bench 2
  • LMArena WebDev
  • METR time horizons
  • Terminal-Bench 4.0
  • Aider Polyglot
  • 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 14.4% 10.1 to 18.7 APEX CC-BY-4.0 source
OSWorld agents 63.3% not published OSWorld (XLANG Lab) none stated General model / 100 steps source
ALE-Bench coding 822 not published ALE-Bench (Sakana AI with AtCoder) CC-BY-4.0 source
SWE-bench Verified (Epoch's own run) coding 73.8% 69.8 to 77.7 Epoch AI CC-BY-4.0 source
ARC-AGI-2 reasoning 11.8% not published ARC Prize CC-BY-4.0 source
Humanity's Last Exam reasoning 24.4% 20.8 to 27.9 Humanity’s Last Exam (CAIS / Scale AI) CC-BY-4.0 source

Configurations rolled up: 2. Rule: median observed configuration per benchmark (lower median, always a real measurement). Harnesses seen: general-model.

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.