BlogLLM Benchmark Scores as Published, Side by Side
Gemma 4 26B A4B: Benchmark Scores as Published
Every result the catalog holds for Gemma 4 26B A4B: 63 tests in 13 areas, from 4 pages. 15 printed by Google itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.
Developer GoogleReleased 2026-04-02Context window 256KInput Text, imageOutput TextWeights Open (Apache 2.0)Data version 2026103001, as of 2026-10-30
Knowledge Work
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GDPval-AA v2 | 807 | NVIDIA, NVIDIA Nemotron 3.5 Lightning 30B-A3B model cardchecked 2026-10-19 | - |
| GDPval-AA v2 (printed as a percentage) | 15.33% | NVIDIA, NVIDIA Nemotron 3.5 Bolt 74B-A9B model cardchecked 2026-10-19 | - |
Agentic Coding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| SWE-bench Verified | 17.4% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15; printed on 1 more page | Harness: internal scaffold; Context: 200000 |
| SWE-Bench Pro | 13.8% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | Harness: internal scaffold; Context: 200000 |
| Terminal-Bench 2.1 | 37.22% | NVIDIA, NVIDIA Nemotron 3.5 Bolt 74B-A9B model cardchecked 2026-10-19; printed on 1 more page | - |
| SWE-bench Multilingual | 17.3% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15; printed on 1 more page | Harness: internal scaffold; Context: 200000 |
| Terminal-Bench 2.0 | 34.2% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | Harness: Harbor/Terminus-2; Context: 256000 |
| NL2Repo-Bench | 11.6 | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | Harness: Claude Code (baselines) |
Science and Math
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPQA Diamond | 82.3% | Google, Gemma 4 model cardchecked 2026-10-15; printed on 3 more pages | - |
| SciCode | 40.28% | NVIDIA, NVIDIA Nemotron 3.5 Bolt 74B-A9B model cardchecked 2026-10-19; printed on 1 more page | - |
| HMMT February 2025 | 91.7% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
| IMO-AnswerBench | 74.3% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
| AIME 2026 | 88.3% | Google, Gemma 4 model cardchecked 2026-10-15; printed on 1 more page | Tools: none |
| HMMT November 2025 | 87.5% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
| HMMT February 2026 | 79% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
Long Context
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| AA-LCR | 57.56% | NVIDIA, NVIDIA Nemotron 3.5 Bolt 74B-A9B model cardchecked 2026-10-19; printed on 1 more page | - |
| MRCR v2 8-needle 128K | 44.1% | Google, Gemma 4 model cardchecked 2026-10-15 | Context: 128k (average) |
Computer Use
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| WebArena (vision) | 38.2% | NVIDIA, NVIDIA Nemotron 3.5 Bolt 74B-A9B model cardchecked 2026-10-19 | - |
Multimodal Understanding
General Knowledge and Reasoning
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Humanity's Last Exam (no tools) | 8.7% | Google, Gemma 4 model cardchecked 2026-10-15; printed on 3 more pages | Tools: none |
| MMLU-Pro | 82.6% | Google, Gemma 4 model cardchecked 2026-10-15; printed on 2 more pages | - |
| Humanity's Last Exam (with tools) | 17.2% | Google, Gemma 4 model cardchecked 2026-10-15 | Tools: search |
| MMLU-Redux | 92.7% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
| C-Eval | 82.5% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
| AA-Omniscience Accuracy | 22.17% | NVIDIA, NVIDIA Nemotron 3.5 Bolt 74B-A9B model cardchecked 2026-10-19; printed on 1 more page | - |
| SuperGPQA | 61.4% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
| AA-Omniscience Non-Hallucination | 14.22% | NVIDIA, NVIDIA Nemotron 3.5 Bolt 74B-A9B model cardchecked 2026-10-19 | - |
| BIG-Bench Extra Hard | 64.8% | Google, Gemma 4 model cardchecked 2026-10-15 | - |
Code Generation
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| LiveCodeBench v6 | 77.1% | Google, Gemma 4 model cardchecked 2026-10-15; printed on 1 more page | - |
| Codeforces rating | 1718 | Google, Gemma 4 model cardchecked 2026-10-15 | - |
| QwenWebBench rating | 1178 | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
Tool Use and Workflows
Web Search and Research
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| BrowseComp | 26.3% | NVIDIA, NVIDIA Nemotron 3.5 Bolt 74B-A9B model cardchecked 2026-10-19; printed on 1 more page | - |
| WideSearch (metric not stated) | 38.3 | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
Instruction Following
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| IFBench | 77.25 | NVIDIA, NVIDIA Nemotron 3.5 Lightning 30B-A3B model cardchecked 2026-10-19 | - |
Multilingual
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MMMLU | 86.3% | Google, Gemma 4 model cardchecked 2026-10-15 | - |
Healthcare
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| MedXpertQA MM | 58.1% | Google, Gemma 4 model cardchecked 2026-10-15 | - |
More from Google
Every number here is read from the same catalog file as the benchmark table, data version 2026103001. Where one page printed the model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.