whatmodel.sh / Gemini 4 Carbon
Gemini 4 Carbon
↗ share
+ contribute
Gemini 4 Carbon is a closed-weights language model from Google, released October 2026.
Limited preview · broad public access is not yet available. Access announcement ↗
Catalogue identity notes
Limited rollout to selected cyber defenders and trusted partners. No broadly available API or weights verified on 2026-10-29; announced future price is not a purchasable API route.
Model information sources
Score within the LMArena (text) comparison set only.
$ how it compares
The LLMs ranked either side of Gemini 4 Carbon, out of 884 with a score, then more from Google.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 2 |  Claude Opus 5.5 | 85±1 | api $8.00/M | ||
| 3 |  GPT-6.1 Terra | 84±1 | api $2.00/M | ||
| 4 | Gemini 4 Carbon | 83±1 | limited preview | ||
| 5 | GPT-6.1 Sol | 83±1 | api $4.00/M | ||
| 6 | Claude Fable 5.1 | 82±1 | api $20.00/M | ||
| more from Google | |||||
| 9 | Gemini 4 Argon | 81±1 | api $4.00/M | ||
| 26 | Gemini 3.8 Flash | 76±1 | api $1.50/M | ||
| 36 | Gemini 3.7 Flash | 75±1 | api $1.50/M | ||
$ what it's good at
agent tasks▁▁▃▆█▄▅▅▄▄▂▃ #2 of 312 * · 46% of its score · 1 test
Beats GPT-6 Astra, GPT-6.1 Terra here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena Agent (October 2026 aggregate IPS) | 0.1604 IPS ✎ | ▁▁▁▃▃▃█▁▁▂▃▁ beats 53/54 | 46% |
coding▁▁▁▃▅▇█▇▅▄▃▃ #3 of 468 * · 34% of its score · 1 test
Beats GPT-6.1 Terra here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena WebDev (Code Arena) | 1768.2 ✎ | ▁▁▃▂▅█▃▆▅▅▁▃ beats 126/129 | 34% |
chat & writing▁▁▁▃▆▄▆█▆▇▇▃ #1 of 401 * · 19% of its score · 1 test
Beats GPT-6 Astra, Claude Opus 5.5 here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena (text) | 1527.5 ✎ | ▁▁▁▂▃▃▃▆▆▇█▃ beats 385/385 | 19% |
other tests2 not in a task · 0.0% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| SEC-Bench Pro (pass@1) lab | 88.2% ✎ | ▄·▄·▄·█▄···▄ beats 6/6 | — |
| ExploitGym (pass@1) lab | 54.8% ✎ | ▄·▄··█·▄▄··▄ beats 6/6 | — |
$ where to run it
No verified API or download recorded yet · add one
$ how the score was worked out · LMArena leaderboard dataset 99%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 LMArena leaderboard dataset | 3 | 99% |
| prior · pull toward the middle | 1% | |
| weighted average | = 83.3 |
not counted (2)
| Google DeepMind Gemini model page and model cards (lab-reported scores) | SEC-Bench Pro (pass@1) [correct] | 88.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| Google DeepMind Gemini model page and model cards (lab-reported scores) | ExploitGym (pass@1) [correct] | 54.8% | fewer than three models have this test yet, or this source ran it for this model alone |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json