THE FORGE RANKINGS·2026-08-30

Gemini 3 Flash Preview

Google DeepMind
21
RANK of 37 ranked
FERROX INDEX65.0 weighted mean of category positions
GRADEC below the field median
EVIDENCE8 / 32 benchmarks measured
QUALITYFULL every category rests on independent benchmarks

Not separable from Kimi K2.6 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

21of 37 measured
37.5
13.1median 38.963.0

Coding

7of 37 measured
57.2
22.4median 50.070.7

Reasoning

26of 37 measured
50.2
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

12/35
1367 best 2177 · GPT-5.6 Sol
ALE-Bench (Sakana AI with AtCoder) CC-BY-4.0

APEX Agents

17/28
24.0% best 45.0% · Claude Fable 5
APEX CC-BY-4.0

FrontierMath

9/21
35.6% best 47.6% · GPT-5.4 (2026-03-…
Epoch AI CC-BY-4.0

SWE-bench Verified (Epoch's own run)

7/21
75.4% best 83.5% · Claude Opus 4.7
Epoch AI CC-BY-4.0

Not measured

24/32
  • LMArena Agent
  • LMArena Text (style-controlled)
  • DeepSWE
  • FrontierCode
  • Humanity's Last Exam
  • LMArena WebDev
  • METR time horizons
  • Terminal-Bench 4.0
  • 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
APEX Agents agents 24.0% 17.5 to 30.5 APEX CC-BY-4.0 source
Terminal-Bench 2 agents 51.0% 45.1 to 56.9 Terminal-Bench v2 Leaderboard CC-BY-4.0 Gemini CLI source
ALE-Bench coding 1367 not published ALE-Bench (Sakana AI with AtCoder) CC-BY-4.0 source
SWE-bench Verified (Epoch's own run) coding 75.4% 71.6 to 79.3 Epoch AI CC-BY-4.0 source
ARC-AGI-2 reasoning 3.3% not published ARC Prize CC-BY-4.0 source
FrontierMath reasoning 35.6% 30.1 to 41.2 Epoch AI CC-BY-4.0 source
GPQA Diamond (Epoch's own run) reasoning 83.2% 79.5 to 86.9 Epoch AI CC-BY-4.0 source
OTIS Mock AIME 2024-2025 reasoning 92.8% 85.9 to 99.7 Epoch AI CC-BY-4.0 source

Configurations rolled up: 4. Rule: median observed configuration per benchmark (lower median, always a real measurement). Harnesses seen: gemini-cli, junie-cli.

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.