Benchmark Almanac

DrivingBench v2

Gemini 4 Carbon 89.5% leads since November 2026
Not in model cards
Version 2

Vision-language models drive a real Tesla through two unseen cone courses with steering, accelerator and brake tools. The score is the furthest progress over three attempts in one conversation, averaged over six trials.

Released November 16, 2026. Progress comes from GPS, accurate to about 1 to 2 m, and doesn’t show whether the car stayed inside the cones. Fine-tuned driving models (VLAs and WAMs) are ranked separately on the site. A human driver using the same tools reached 95%.

Released
November 2026
Kind
Real robots
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images, Time series → Actions
Interaction
Agentic
Run by
A host
Based on
DrivingBench
In views
Agents, Vision, Robotics
Strongest model tested
Claude Fable 5.2 #2 on the capability index
Last model added
November 2026
Scores refreshed
November 18th, 2026
8
Models scored on its own leaderboard

Gemini 4 Carbon leads with 89.5%

Best score per model from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

  1. 1 Gemini 4 Carbon best of three attempts 89.5%
  2. 2 GPT-6 Astra best of three attempts 35.3%
  3. 3 Claude Opus 5.5 best of three attempts 30.5%
  4. 4 Claude Fable 5.2 best of three attempts 21.3%
  5. 5 GPT-6.1 Sol best of three attempts 21%
  6. 6 Claude Fable 5.1 best of three attempts 12.4%
  7. 7 Grok 4.6 best of three attempts 10.8%
  8. 8 GPT-5.6 Sol best of three attempts 9.40%

Source: Official DrivingBench v2 leaderboard

#ModelScoreSettingDateSource
1 Gemini 4 Carbon 89.5% best of three attempts Oct 2026 Official leaderboard
2 GPT-6 Astra 35.3% best of three attempts Sep 2026 Official leaderboard
3 Claude Opus 5.5 30.5% best of three attempts Sep 2026 Official leaderboard
4 Claude Fable 5.2 21.3% best of three attempts Nov 2026 Official leaderboard
5 GPT-6.1 Sol 21% best of three attempts Sep 2026 Official leaderboard
6 Claude Fable 5.1 12.4% best of three attempts Sep 2026 Official leaderboard
7 Grok 4.6 10.8% best of three attempts Aug 2026 Official leaderboard
8 GPT-5.6 Sol 9.40% best of three attempts Jul 2026 Official leaderboard

Rankings by part

We read 2 parts from the benchmark’s own leaderboard: Course A and Course B. Each ranks models on one part of the test. Gemini 4 Carbon leads both. Arrows show how far a model moved from its overall rank among the models we track.

Course A

The original layout: a left, a long right sweep, a left hairpin and a sharp right into the finish. 8 models we track have a score here.

  1. 1 Gemini 4 Carbon 100%
  2. 2 Claude Opus 5.5 ↑1 38.2%
  3. 3 GPT-6 Astra ↓1 33.8%
  4. 4 Claude Fable 5.2 32.6%
  5. 5 GPT-6.1 Sol 18.3%
  6. 6 Claude Fable 5.1 15.8%
  7. 7 GPT-5.6 Sol ↑1 10.9%
  8. 8 Grok 4.6 ↓1 10.7%

The best score, 89.5%, costs $1.66 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Google DeepMind Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.10 $0.20 $0.50 $1.00 $2.00 $5.00 $10 Cost per task, log scale Gemini 4 Carbon, best of three attempts GPT-6.1 Sol, best of three attempts 0% 20% 40% 60% 80% 100% $0.10 $1.00 $10 Cost per task, log scale Gemini 4 Carbon, best of three attempts GPT-6.1 Sol, best of three attempts

Source: Official DrivingBench v2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Carbon best of three attempts 89.5% $1.66 66,000
GPT-6 Astra best of three attempts 35.3% $3.75 4,000
Claude Opus 5.5 best of three attempts 30.5% $2.27 12,000
Claude Fable 5.2 best of three attempts 21.3% $3.41 17,000
GPT-6.1 Sol best of three attempts 21% $0.25 3,000
Claude Fable 5.1 best of three attempts 12.4% $3.94 25,000
Grok 4.6 best of three attempts 10.8% $0.62 6,000
GPT-5.6 Sol best of three attempts 9.40% $0.61 4,000

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind
Own leaderboard Passed 85% in Nov 2026, by Gemini 4 Carbon

Source: Official DrivingBench v2 leaderboard

Gemini 4 Carbon on the official leaderboard passed 85% in October 2026

Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Gemini 4 Carbon GPT-6 Astra Claude Fable 5.1 GPT-5.6 Sol 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Gemini 4 Carbon GPT-6 Astra

Source: Official DrivingBench v2 leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic OpenAI Google DeepMind
DrivingBench v2 Passed 85% in Nov 2026, by Gemini 4 Carbon
DrivingBench Passed 85% in Sep 2026, by GPT-6 Astra
DrivingBench v2 Passed 85% in Nov 2026, by Gemini 4 Carbon

Source: Official leaderboards

More in Robotics

All 172 in Robotics
  • ERQA In use 400 multiple-choice visual questions from robot and egocentric footage on spatial, trajectory, action and state reasoning. Accuracy. Gemini Robotics ER 2 78.5%
  • SimplerEnv In use Real-to-sim replicas of Google Robot and WidowX/Bridge setups. Policies scored by success rate on pick, move, drawer and Bridge tasks. Praxis-1 80%
  • RoboLab In use Zero-shot DROID Franka policies in Isaac Lab on 120 tabletop tasks at three instruction levels, 10 episodes each. Success rate. Praxis-1 46.2%
  • RoboDojo In use Generalist manipulation policies run 42 Isaac Sim tasks and 18 remote real-robot tasks. Scored by task score and success rate. GPT-6 Astra 27.6%
  • Progress classification (Google DeepMind) In use Models assign each frame of a robot video to one of five task-progress bands from 0 to 100 percent. Scored by accuracy. Gemini Robotics ER 2 57.4%
  • Human proximity (Google DeepMind robotics) In use Models judge from robot camera views whether a human is within 1 meter, a spatial awareness check for safe robot behavior. Scored by accuracy. Gemini Robotics ER 2 93%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (8 entries).
  • Pass Models scored 8 models we track.
  • Pass Newest score The newest is from November 2026, for Claude Fable 5.2.
  • Pass Frontier models 6 of the top 10 have a score, led by Claude Fable 5.2 (#2).
  • Note Headroom Saturated. The best is 89.5%, past 85% since October 2026.
Leaderboard import 3 pass
  • Pass Importer We read the board on drivingbench.com: 8 rows for models we track at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check