Google model product
Gemini 4 Carbon is Google DeepMind's model, building on Gemini 4 Argon with gains in vulnerability finding, exploit validation, patching, and long-horizon agentic coding.
Updated Nov 7, 2026. Default version: Gemini 4 Carbon
This profile uses the model's current scored version. Arena ratings and prices are shown separately.
Benchmark scores for Gemini 4 Carbon.
Benchmark | Score | Rank | Participants | Percentile | Evidence | Evaluated |
|---|
| BenchmarkBlueprint-Bench 2 | Score58.60% | Rank01 | Participants3 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkCWE-Bench | Score71.00% | Rank01 | Participants3 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkExploitGym | Score53.50% | Rank01 | Participants10 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkGraphwalks BFS >128k | Score96.20% | Rank01 | Participants13 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkSEC-bench Pro | Score88.20% | Rank01 | Participants8 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkTerminal-Bench-Science 0.1 | Score64.80% | Rank01 | Participants10 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkVending-Bench 2 | Score16,484.00 usd | Rank01 | Participants5 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkWebDev Arena | Score1,768.00 points | Rank01 | Participants2 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkLM Arena Agent Leaderboard | Score16.72% | Rank01 | Participants44 | Percentile100.00% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Praise Complaint | Score40.12% | Rank01 | Participants44 | Percentile100.00% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Steerability | Score19.32% | Rank01 | Participants44 | Percentile100.00% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Task Outcome Explicit | Score18.02% | Rank01 | Participants44 | Percentile100.00% | EvidenceA | Evaluated |
| BenchmarkLM Arena Text | Score1,540.04 rating | Rank01 | Participants220 | Percentile100.00% | EvidenceA | Evaluated |
| BenchmarkLM Arena Text Factuality | Score1,521.59 rating | Rank01 | Participants133 | Percentile100.00% | EvidenceA | Evaluated |
| BenchmarkLM Arena Text Style Control | Score1,527.51 rating | Rank01 | Participants220 | Percentile100.00% | EvidenceA | Evaluated |
| BenchmarkChartography (no tools) | Score72.50% | Rank01 | Participants4 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkOSWorld 2.1 (partial) | Score81.50% | Rank01 | Participants7 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkRiemannBench | Score80.00% | Rank01 | Participants2 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkVals Index | Score70.40% | Rank01 | Participants2 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkTerminal-Bench 4.0-AA | Score65.20% | Rank01 | Participants5 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkGray Swan IPI Benchmark | Score0.00% | Rank02 | Participants2 | Percentile0.00% | EvidenceC | Evaluated |
| BenchmarkLM Arena Webdev | Score1,768.15 rating | Rank03 | Participants108 | Percentile98.13% | EvidenceA | Evaluated |
| BenchmarkFrontierSWE V2 | Score61.20% | Rank03 | Participants6 | Percentile60.00% | EvidenceC | Evaluated |
| BenchmarkAutomationBench | Score53.90% | Rank04 | Participants30 | Percentile89.66% | EvidenceC | Evaluated |
| BenchmarkLM Arena Agent Tool Hallucination | Score0.51% | Rank19 | Participants44 | Percentile58.14% | EvidenceA | Evaluated |
| BenchmarkLM Arena Agent Bash Reliability | Score-5.50% | Rank34 | Participants44 | Percentile23.26% | EvidenceA | Evaluated |
Preference and agent-evaluation results for the default version.
Arena | Category | Rank | Rating / score | Votes | Observations | Result date |
|---|
| Arenaagent | Categoryoverall | Rank01 | Rating / score0.17 | VotesN/A | Observations7.5M | Result date |
| Arenaagent praise complaint | Categoryoverall | Rank01 | Rating / score0.40 | VotesN/A | Observations10.2K | Result date |
| Arenaagent steerability | Categoryoverall | Rank01 | Rating / score0.19 | VotesN/A | Observations209K | Result date |
| Arenaagent task outcome explicit | Categoryoverall | Rank01 | Rating / score0.18 | VotesN/A | Observations22.7K | Result date |
| Arenatext | Categorychinese | Rank01 | Rating / score1,616.20 | Votes474 | ObservationsN/A | Result date |
| Arenatext | Categorycoding | Rank01 | Rating / score1,569.49 | Votes1,106 | ObservationsN/A | Result date |
| Arenatext | Categoryenglish | Rank01 | Rating / score1,530.29 | Votes1,558 | ObservationsN/A | Result date |
| Arenatext | Categoryexclude ties | Rank01 | Rating / score1,568.81 | Votes3,354 | ObservationsN/A | Result date |
| Arenatext | Categoryexpert | Rank01 | Rating / score1,596.37 | Votes428 | ObservationsN/A | Result date |
| Arenatext | Categoryhard prompts | Rank01 | Rating / score1,552.51 | Votes2,745 | ObservationsN/A | Result date |
| Arenatext | Categoryhard prompts english | Rank01 | Rating / score1,545.11 | Votes1,022 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry business and management and financial operations | Rank01 | Rating / score1,522.01 | Votes730 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry entertainment and sports and media | Rank01 | Rating / score1,526.10 | Votes1,175 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry legal and government | Rank01 | Rating / score1,565.35 | Votes284 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry software and it services | Rank01 | Rating / score1,555.23 | Votes1,621 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry writing and literature and language | Rank01 | Rating / score1,543.18 | Votes1,272 | ObservationsN/A | Result date |
| Arenatext | Categoryinstruction following | Rank01 | Rating / score1,539.19 | Votes1,484 | ObservationsN/A | Result date |
| Arenatext | Categorylonger query | Rank01 | Rating / score1,553.21 | Votes1,882 | ObservationsN/A | Result date |
| Arenatext | Categorymulti turn | Rank01 | Rating / score1,564.68 | Votes602 | ObservationsN/A | Result date |
| Arenatext | Categorynon english | Rank01 | Rating / score1,536.92 | Votes2,845 | ObservationsN/A | Result date |
| Arenatext | Categoryoverall | Rank01 | Rating / score1,540.04 | Votes4,403 | ObservationsN/A | Result date |
| Arenatext | Categoryrussian | Rank01 | Rating / score1,563.81 | Votes647 | ObservationsN/A | Result date |
| Arenatext factuality | Categorycoding | Rank01 | Rating / score1,569.68 | Votes1,106 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryexclude ties | Rank01 | Rating / score1,547.70 | Votes3,354 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryhard prompts | Rank01 | Rating / score1,544.67 | Votes2,745 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry business and management and financial operations | Rank01 | Rating / score1,511.16 | Votes714 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry entertainment and sports and media | Rank01 | Rating / score1,503.55 | Votes1,175 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry software and it services | Rank01 | Rating / score1,548.88 | Votes1,621 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry writing and literature and language | Rank01 | Rating / score1,521.60 | Votes1,272 | ObservationsN/A | Result date |
| Arenatext factuality | Categorylonger query | Rank01 | Rating / score1,538.93 | Votes1,882 | ObservationsN/A | Result date |
Official vendor API pricing appears first, followed by individual provider offers.
The default version has no current input or output token prices.
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
Provider-specific output speed and catalog latency for Gemini 4 Carbon. Runtime does not affect the capability score.
No provider-specific speed or latency record is linked to the default version yet.
Output Speed is generated output tokens received per second. Catalog latency is reported separately from observed provider TTFT.
Technical details for the model's default version.
Available versions of this model. The score column identifies the version used in the overall ranking.
Version | Released | LLMBoard | Parameters | Context | Max output | Open weights | License |
|---|
| VersionGemini 4 Carbon | Released | LLMBoard98.77 | ParametersN/A | ContextN/A | Max outputN/A | Open weightsNo | LicenseProprietary |
Recommendations prioritize the same model type and family, then the closest LLMBoard score.
Key information about Gemini 4 Carbon and its available data.
Gemini 4 Carbon is Google DeepMind's LLM, announced on 2026-10-28, with stated gains over Gemini 4 Argon in vulnerability finding, exploit validation, patching, and long-horizon agentic coding.
It is in a limited rollout to a subset of cyber defenders and, in phases, to trusted partners through Google's Fairwind Program; there is no public API endpoint yet.
Data as of 2026-11-06.
Common questions about Gemini 4 Carbon.
Gemini 4 Carbon's default version was released on Oct 28, 2026.
No official standard PAYG price is currently available for Gemini 4 Carbon.
Gemini 4 Carbon was created by Google.
A context window is not available for the default version.
No. The default version is not marked as having publicly available weights.
No provider offering is currently linked to the default version.