Natural2Code
DiffusionGemma 26B A4B 94% leads since July 2026 Models write Python functions in HumanEval format from new prompts kept off the web. Unit tests decide correctness.
In-house Google DeepMind set that was never released, built to avoid web leakage.
- Made by
- Google DeepMind
- Released
- December 2023
- Kind
- Function-level problems
- Level
- School
- Scored
- Checked automatically
- Answers
- Code
- Tasks
- Kept private
- Strongest model tested
- Gemma 3 27B #164 on the capability index
- Last in a launch post
- March 2025, dropped after it saturated
- Last model added
- July 2026
DiffusionGemma 26B A4B leads with 94% among models released by June 2026
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in June 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
- Gemini 4 Carbon
- GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 1 DiffusionGemma 26B A4B 94%
- 2 Gemini 2.0 Flash Dec 2024 92.9%
- 3 Gemma 3 27B 84.5%
- 4 Gemini 1.5 Pro May 2024 82.6%
- 5 Gemma 3 12B 80.7%
- 6 Gemini 1.5 Pro Feb 2024 77.7%
- 7 Gemini 1.0 Ultra 74.9%
- 8 Gemma 3 4B 70.3%
- 9 Gemma 3 1B 56%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | DiffusionGemma 26B A4B | 94% | Jul 2026 | Lab model card | |
| 2 | Gemini 2.0 Flash Dec 2024 | 92.9% | Dec 2024 | Lab model card | |
| 3 | Gemma 3 27B | 84.5% | Mar 2025 | Lab model card | |
| 4 | Gemini 1.5 Pro May 2024 | 82.6% | May 2024 | Lab model card | |
| 5 | Gemma 3 12B | 80.7% | Mar 2025 | Lab model card | |
| 6 | Gemini 1.5 Pro Feb 2024 | 77.7% | Feb 2024 | Lab model card | |
| 7 | Gemini 1.0 Ultra | 74.9% | Dec 2023 | Lab model card | |
| 8 | Gemma 3 4B | 70.3% | Mar 2025 | Lab model card | |
| 9 | Gemma 3 1B | 56% | Mar 2025 | Lab model card |
5 models have held number one in labs’ own numbers since its release in December 2023
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Labs’ model cards
Google DeepMind’s own number for Gemini 2.0 Flash passed 85% in December 2024
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
Labs dropped it from launch posts once the best score passed 85%
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| Mar 12, 2025 | Gemma 3 27B | 84.5 | Natural2Code 0-shot (IT) |
| Dec 11, 2024 | Gemini 2.0 Flash | 92.9 | Natural2Code |
| May 14, 2024 | Gemini 1.5 Pro | 82.6 | Natural2Code chat preamble, 0-shot |
| Feb 15, 2024 | Gemini 1.5 Pro | 77.7 | Natural2Code chat preamble, 0-shot |
| Dec 6, 2023 | Gemini 1.0 Ultra | 74.9 | Natural2Code, 0-shot |
5 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Jul 31, 2026 | DiffusionGemma 26B A4B technical report | 94 | Natural2Code (internal), TD, 90.1 no-think (Table 3) | Capabilities |
| Mar 25, 2025 | Gemma 3 4B technical report | 70.3 | N2C (Natural2Code), IT, 0-shot pass@1 (Table 18) | Capabilities |
| Mar 25, 2025 | Gemma 3 27B technical report | 84.5 | N2C (Natural2Code), IT, 0-shot pass@1 (Table 18) | Capabilities |
| Mar 25, 2025 | Gemma 3 1B technical report | 56 | N2C (Natural2Code), IT, 0-shot pass@1 (Table 18) | Capabilities |
| Mar 25, 2025 | Gemma 3 12B technical report | 80.7 | N2C (Natural2Code), IT, 0-shot pass@1 (Table 18) | Capabilities |
More in Coding
- SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
- LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests.
DeepSeek-V4-Pro 93.5% - SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
- DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
- SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases.
Seed2.1 Pro 59.8% - Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471