Gemini 4 Carbon

Leaderboards

Where Gemini 4 Carbon sits among the top 20 in each category.

By task

Scored from votes on each kind of prompt alone, on the same 1,000-point scale.

Test results

Every test Gemini 4 Carbon has a published result on, out of 100%, with its rank among the models that took it.

• AutomationBench

  #4 / 27

  AutomationBench is a tool-use benchmark that evaluates AI agents on automating real-world workflows, testing their ability to orchestrate tools and complete multi-step automation tasks.

  53.9%

  Self-reported · In the agents score

• Graphwalks BFS >128k

  #1 / 13

  A graph reasoning benchmark that evaluates language models' ability to perform breadth-first search (BFS) operations on graphs with context length over 128k tokens, testing long-context reasoning capabilities.

  96.2%

  Self-reported

• ExploitGym

  #1 / 10

  ExploitGym is a large-scale, realistic benchmark built from real-world vulnerabilities across userspace programs, Google's V8 engine, and the Linux kernel. Given a vulnerability and a proof-of-vulnerability input, agents must craft a working end-to-end exploit that achieves unauthorized code execution.

  53.5%

  Self-reported

• Terminal-Bench-Science 0.1

  #1 / 10

  Terminal-Bench-Science 0.1 is a Stanford-led community benchmark of 70 tasks drawn from scientific research workflows across the life, physical, earth, mathematical, and engineering sciences. Tasks are authored and reviewed by scientists and researchers; scores are typically reported as accuracy with relatively large standard error due to strongly bimodal per-task outcomes.

  64.8%

  Self-reported

• SEC-bench Pro

  #1 / 8

  SEC-bench Pro is a self-evolving software-security benchmark that measures agent bug hunting on critical, high-complexity systems. It instantiates validated vulnerabilities across the V8 and SpiderMonkey JavaScript engines as reproducible vulnerability-discovery and proof-of-concept-generation tasks with oracle-based validation.

  88.2%

  Self-reported

• OSWorld 2.1 (partial)

  #1 / 7

  Partial-credit computer-use score on OSWorld 2.1, as reported in the Claude Sonnet 5.5 release. Separate from OSWorld 2.0 and strict completion scores.

  81.5%

  Self-reported

• FrontierSWE V2

  #3 / 6

  FrontierSWE V2 evaluates agents on open-ended technical projects and reports mean@5. Kept separate from FrontierSWE V1 dominance-score results, which are not comparable.

  61.2%

  Self-reported

• Terminal-Bench 4.0-AA

  #1 / 5

  Artificial Analysis reported accuracy on terminal tasks with executable final-container-state verifiers.

  65.2%

  Self-reported

• Vending-Bench 2

  #1 / 5

  Vending-Bench 2 tests longer horizon planning capabilities by evaluating how well AI models can manage a simulated vending machine business over extended periods. The benchmark measures a model's ability to maintain consistent tool usage and decision-making for a full simulated year of operation, driving higher returns without drifting off task.

  1648400.0%

  Self-reported

• Chartography (no tools)

  #1 / 4

  Chartography visual chart recognition evaluated without tools, distinct from tool-assisted results.

  72.5%

  Self-reported

• Blueprint-Bench 2

  #1 / 3

  Blueprint-Bench 2 is an agentic spatial reasoning benchmark that evaluates a model's ability to understand, plan, and reason over architectural blueprints and other structured spatial documents. Scores are reported as a normalized score.

  58.6%

  Self-reported

• CWE-Bench

  #1 / 3

  CWE-Bench evaluates autonomous agents on finding and patching real-world software vulnerabilities.

  71.0%

  Self-reported

• Gray Swan IPI Benchmark

  #2 / 2

  The Gray Swan IPI Benchmark evaluates indirect prompt-injection robustness by measuring attack success within a fixed number of attempts; lower is better.

  0.0%

  Unverified

• RiemannBench

  #1 / 2

  RiemannBench evaluates models on advanced mathematics and scientific reasoning tasks, as reported on the Surge public leaderboard.

  80.0%

  Unverified

• Vals Index

  #1 / 2

  Vals Index measures economic impact across finance, coding, legal, and tax work, with every sector weighted by its contribution to U.S. GDP.

  70.4%

  Unverified

• WebDev Arena

  #1 / 2

  WebDev Arena is a leaderboard for evaluating AI models on web development tasks, including zero-shot generation, complex prompts, and interactive web UI creation. Models are ranked using Elo ratings based on their performance in coding and web development challenges.

  88.4%

  Unverified