Gemini 4 Carbon
Released
2026-10-28
Last refreshed
2026-10-28
Status
Researched 1d ago
ProprietaryCommercial use: restrictedMultimodalCodingAgentsVisionCodingAgentMultimodal
Gemini 4 Carbon is a released coding, agents, and vision model; evaluate it while provider pricing coverage matures.
Google DeepMind releases · 51 in the last 12 months · this family litChangelog →
Specifications
- Family
- Gemini 4
- Released
- 2026-10-28
- Specialization
- general
- Openness
- Proprietary
- License
- ProprietaryRestricted
- Weights
- Not released
- Code
- Unknown
- Training
- Pretrained
Created by
Pricing
No tracked provider pricing is available yet.
Links
About
Gemini 4 Carbon is a proprietary model in the Gemini 4 family. The structured metadata tracks multimodal input and reasoning. Headline tracked benchmarks include SEC-Bench Pro 88.2, ExploitGym 54.8, and AutomationBench 53.9.
Providers
No active provider routes are tracked for this model yet.
Capabilities
VisionMultimodalReasoning
Benchmark peer barsfor Coding
No task-mapped benchmark peers are available for this model yet.
Benchmark scores(5)
Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
| Benchmark | Score | Version | Evaluation | Source |
|---|---|---|---|---|
| 88.2 | SEC-bench Pro, 183-task May 2026 JavaScript engine splitObserved 2026-10-28 | Configuration: Gemini 4 Carbon (highest thinking), Antigravity harness Harness: Antigravity Evaluator: Google DeepMind Confidence: self-reported Notes: Pass@1 averaged over 3 independent attempts per task; blog also states 88.2%. | ||
| 54.8 | ExploitGym (open-source setup)Observed 2026-10-28 | Configuration: Gemini 4 Carbon (highest thinking), Antigravity harness, single attempt, no internet, mitigations off Harness: Antigravity Evaluator: Google DeepMind Confidence: self-reported Notes: Methodology measured both the 6-hour default and a 10-hour extended timeout; the results table does not say which timeout 54.8% uses. Blog also states 54.8%. | ||
| 53.9 | AutomationBench private setObserved 2026-10-28 | Configuration: Gemini 4 Carbon Evaluator: Google DeepMind Confidence: self-reported Notes: Google states the result is sourced from the official Zapier public leaderboard; Argon on the same table is 51.3%, matching the existing Argon row. | ||
| 58.6 | Blueprint-Bench 2 (similarity score)Observed 2026-10-28 | Configuration: Gemini 4 Carbon Evaluator: Google DeepMind Confidence: self-reported Notes: Google states the result is reported from the Andon Labs official public leaderboard. | ||
| 1768.0 | webdev overall 2026-10-28Observed 2026-10-29 | Configuration: Gemini 4 Carbon (High) Harness: LMArena blind pairwise human votes, Bradley-Terry Evaluator: LMArena Confidence: confirmed Notes: LMArena WebDev leaderboard, webdev category, published 2026-10-28 (HF lmarena-ai/leaderboard-dataset @4491b06c, latest split). Board entry 'Gemini 4 Carbon (High)': rank 4, Arena Score 1768.2 (95% CI 1753.2-1783.2), 2230 votes. Google blog states 1768. Retrieved 2026-10-29. | Source Source: LMArena (lmarena.ai), CC BY 4.0 |
Migration checks
No linked migration route is available for this model yet.
Created by
Pricing
No tracked provider pricing is available yet.
Links