Gemma 4 26B A4B IT — Benchmarks

Benchmark scores for Gemma 4 26B A4B IT aggregated from public leaderboards (10 boards), with how it ranks among open models and how it compares with its nearest neighbours. See hardware requirements for what you need to run it.

llmrun Composite Score →

105.8

90% range 102.0–109.6 · from 8 boards

Coding103.690% range 95–112
AgentsNot enough data yet
MathNot enough data yet
Science102.290% range 95–110
Reasoning102.090% range 94–110

Gemma 4 26B A4B IT reasoning benchmarks

Multi-step logic and problem-solving — can the model think through a hard problem rather than recall an answer.

GPQA Diamond
GPQA Diamond is a set of extremely hard, graduate-level science questions (physics, chemistry, biology) written by domain experts and filtered so that skilled non-experts with web access still fail. It measures genuine reasoning rather than memorization.
73.2%
73.2%
median 54.0%open leader 93.1%
Overall best: Gemini 4 Argon (high) 96.4%
70th percentile · #30 of 98 open · #163 of 300 all
#30 of 98
#163 of 300
Independent
Epoch AI · Apr 2026
CritPt
CritPt (Critical Points) tests graduate-level physics problem-solving on research-style questions spanning mechanics, electromagnetism, quantum mechanics, and more. It was created by physicists to probe whether a model can reason through multi-step physics rather than pattern-match textbook answers.
0.0%
0.0%
median 1.1%open leader 26.6%
Overall best: GPT 5.6 Sol Max 32.3%
15th percentile · #51 of 66 open · #188 of 219 all
#51 of 66
#188 of 219
Lab-reported
Epoch AI · Apr 2026
DTBench
DTBench is a set of handcrafted, expert-validated multiple-choice questions on decision theory — Newcomb-like problems and related puzzles — built by the Conceptual Reasoning Index project with Anthropic. It tests whether a model can follow an abstract argument rather than recall a fact.
74.9%
74.9%
median 67.7%open leader 93.9%
Overall best: Claude Opus 5.5 Max 98.9%
58th percentile · #29 of 68 open · #132 of 214 all
#29 of 68
#132 of 214
Lab-reported
Epoch AI · Apr 2026
LMCA
LMCA (Language Model Conceptual Argumentation) scores models on expert-rated arguments across philosophy, decision theory, and AI risk — domains where there is no empirical feedback loop to check an answer against. It is part of the Conceptual Reasoning Index, built with Anthropic by researchers Emery Cooper and Caspar Oesterheld.
29.7%
29.7%
median 24.7%open leader 55.5%
Overall best: Claude Opus 5.5 Max 68.2%
64th percentile · #20 of 54 open · #117 of 176 all
#20 of 54
#117 of 176
Lab-reported
Epoch AI · Apr 2026

Gemma 4 26B A4B IT coding benchmarks

Writing, editing, and fixing real code — measured by whether the resulting program actually runs and passes tests.

SciCode
SciCode asks a model to write real scientific-computing code — implementing multi-step numerical methods drawn from physics, chemistry, biology, and math research — and checks the output against reference solutions. It was built by a team of scientists and ML researchers to test coding ability on genuine research problems, not typical software tasks.
40.1%
40.1%
median 38.7%open leader 60.9%
Overall best: Claude Opus 5.5 Max 66.9%
55th percentile · #30 of 66 open · #160 of 207 all
#30 of 66
#160 of 207
Lab-reported
Epoch AI · Apr 2026
WeirdML
WeirdML asks models to write machine-learning code for small, deliberately unusual datasets, then runs the code and scores the accuracy it actually achieves. Built by Håvard Tveit Ihle, it rewards practical problem-solving over recognizing a familiar task.
35.2%
35.2%
median 39.5%open leader 82.6%
Overall best: GPT 6 Astra Promax 93.6%
27th percentile · #34 of 46 open · #131 of 162 all
#34 of 46
#131 of 162
Independent
Epoch AI · Apr 2026
ALE-Bench
ALE-Bench, from Sakana AI and AtCoder, tests long-horizon algorithmic optimisation using problems drawn from AtCoder Heuristic Contests, where there is no single correct answer and solutions are scored on quality. Results are reported as an AtCoder-style performance rating, where higher is better.
927.2
927.2
median 694.1open leader 1524.5
Overall best: GPT 6 Astra Max 2951.3
74th percentile · #12 of 43 open · #59 of 122 all
#12 of 43
#59 of 122
Lab-reported
Epoch AI · Apr 2026

Other results (1)· leaderboard-only, not part of the llmrun Score

LMArena WebDev
Code Arena (formerly WebDev Arena) is Arena's head-to-head coding leaderboard: two models each build a real web app or UI from the same prompt, and people vote blind on which result is better. The rating shown is a relative Bradley-Terry score, not a percentage — compare it only to other models on this board.
1359.5
1359.5
median 1411.4open leader 1655.2
Overall best: Claude Opus 5.5 (Max) 1807.3
33rd percentile · #35 of 52 open · #105 of 138 all
#35 of 52
#105 of 138
Arena
Arena · Oct 2026

Gemma 4 26B A4B IT math benchmarks

Symbolic and quantitative problem-solving, from competition math to multi-step calculation.

AIME 2024/2025
AIME (American Invitational Mathematics Examination) is a prestigious high-school competition of hard, integer-answer problems. It's a widely-cited yardstick for multi-step mathematical reasoning, here on the 2024–2025 papers.
82.2%
82.2%
median 52.8%open leader 98.6%
Overall best: Claude Fable 5 (high) 100.0%
73rd percentile · #24 of 86 open · #111 of 283 all
#24 of 86
#111 of 283
Independent
Epoch AI · Apr 2026

Gemma 4 26B A4B IT instruction following benchmarks

How precisely a model obeys explicit, verifiable instructions — format, length, and content constraints — and, more broadly, how much people prefer its responses head-to-head in blind human voting.

LMArena Text
Arena (arena.ai, formerly LMArena / Chatbot Arena) ranks models by millions of blind, head-to-head human votes on real prompts. The Arena Score shown here is a Bradley-Terry rating computed with Arena's default style control, so higher means people preferred that model's answers more often — it's a relative rating, not a percentage, so compare it only to other models on this board.
1437.5
1437.5
median 1321.5open leader 1487.1
Overall best: Gemini 4 Argon (High) 1523.9
88th percentile · #23 of 187 open · #100 of 405 all
#23 of 187
#100 of 405
Arena
Arena · Oct 2026

Score vs VRAM in the 24 GB tier

Gemma 4 26B A4B IT
Other models
Best Score for its VRAM
x: VRAM (GB) · y: llmrun Score

#5 of 35 that fit 24 GB

Each dot is a model at the best quant that fits a 24 GB GPU. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.

Full 24 GB ranking →

Gemma 4 26B A4B IT vs its neighbours

Same boards for the nearest open models per llmrun Score and the best model that fits 24 GB. The highest score in each row is marked.

Board	Gemma 4 26B A4B ITthis model	Qwen3.8 27Bbest in classQ4_K_M · 18 GB	Gemma 4 31B ITnearest by ScoreQ4_K_M · 22 GB	Qwen3.6 35B A3Bnearest by ScoreQ4_K_M · 22 GB	Qwen3.5 35B A3Bnearest by ScoreQ4_K_M · 22 GB
llmrun Score	105.8	122.1	106.8	106.8	105.8
GPQA Diamond	73.2%	—	75.8%	84.9%	83.5%
CritPt	0.0%	5.4%	1.4%	0.3%	0.6%
DTBench	74.9%	88.0%	82.7%	73.9%	80.0%
LMCA	29.7%	41.4%	39.3%	29.7%	29.5%
SciCode	40.1%	46.6%	45.5%	36.6%	29.3%
WeirdML	35.2%	—	52.3%	34.5%	—
ALE-Bench	927.2	—	925.5	—	—
AIME 2024/2025	82.2%	—	73.3%	86.7%	70.0%
LMArena Text	1437.5	1437.8	1452.2	—	1394.0
Compare head to head		Compare	Compare	Compare	Compare

Scores aggregated from public benchmark sources (Epoch AI: 8 boards, Arena: 2 boards) — each linked from the benchmark pages. llmrun does not run these benchmarks. Percentiles and ranks are among open models in the llmrun catalogue; the median and leader marks refer to the same field.