Benchmark Almanac

GeoDrive-Bench

GPT-5.4 91.9% leads since June 2026
Not in model cards

Models choose driving actions from road images and local traffic conventions. The benchmark covers six countries and omits explicit country labels.

Made by
Yingzi Ma and colleagues
Released
June 2026
Kind
Embodied reasoning
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Vision
Strongest model tested
GPT-5.4 #28 on the capability index
Last model added
March 2026
Scores refreshed
November 20th, 2026
9
Models scored on its own leaderboard

GPT-5.4 leads with 91.9% among models released by March 2026

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

The newest model with a score here came out in March 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-5.4 Rule-Given 91.9%
  2. 2 Qwen3-VL-235B-A22B Rule-Given 90.8%
  3. 3 InternVL3-8B Rule-Given 81.3%
  4. 4 Qwen3-VL-8B-Instruct Rule-Given 80.2%
  5. 5 Gemma 3 12B Rule-Given 76%
  6. 6 Qwen2.5-VL-7B Rule-Given 73.1%
  7. 7 Llama 3.2 11B Rule-Given 68.9%
  8. 8 llava-v1.6-mistral-7b Rule-Given 62.8%
  9. 9 InternVL3.5-8B Rule-Given 53.8%

Source: Official GeoDrive-Bench leaderboard

#ModelScoreSettingDateSource
1 GPT-5.4 91.9% Rule-Given Mar 2026 Official leaderboard
2 Qwen3-VL-235B-A22B 90.8% Rule-Given Sep 2025 Official leaderboard
3 InternVL3-8B 81.3% Rule-Given Apr 2025 Official leaderboard
4 Qwen3-VL-8B-Instruct 80.2% Rule-Given Oct 2025 Official leaderboard
5 Gemma 3 12B 76% Rule-Given Mar 2025 Official leaderboard
6 Qwen2.5-VL-7B 73.1% Rule-Given Jan 2025 Official leaderboard
7 Llama 3.2 11B 68.9% Rule-Given Sep 2024 Official leaderboard
8 llava-v1.6-mistral-7b 62.8% Rule-Given Jan 2024 Official leaderboard
9 InternVL3.5-8B 53.8% Rule-Given Aug 2025 Official leaderboard

Rankings by part

We read 24 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Perception: CN, Perception: US and Perception: UK. GPT-5.4 leads 13 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Perception: CN

9 models we track have a score here.

  1. 1 Qwen3-VL-235B-A22B ↑1 93.2%
  2. 2 GPT-5.4 ↓1 91.1%
  3. 3 InternVL3-8B 75.8%
  4. 4 Qwen2.5-VL-7B ↑2 73.2%
  5. 5 Qwen3-VL-8B-Instruct ↓1 65.8%
  6. 6 Gemma 3 12B ↓1 61.6%
  7. 7 InternVL3.5-8B ↑2 51.6%
  8. 8 Llama 3.2 11B ↓1 46.3%
  9. 9 llava-v1.6-mistral-7b ↓1 27.4%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5.4
Direct Rule-Given
91.9%
Qwen3-VL-235B-A22B
Direct Rule-Given
90.8%
InternVL3-8B
Reasoning Rule-Given
81.3%
Qwen3-VL-8B-Instruct
Reasoning Direct Rule-Given
80.2%
Gemma 3 12B
Direct Rule-Given
76%
Qwen2.5-VL-7B
Reasoning Direct Rule-Given
73.1%
Llama 3.2 11B
Direct Rule-Given
68.9%
llava-v1.6-mistral-7b
Direct Rule-Given
62.8%
InternVL3.5-8B
Reasoning Rule-Given
53.8%

Source: Official GeoDrive-Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.4 Rule-Given 91.9%
Qwen3-VL-235B-A22B Rule-Given 90.8%
GPT-5.4 Reasoning 88%
Qwen3-VL-235B-A22B Reasoning 82.7%
GPT-5.4 Direct 82.5%
InternVL3-8B Rule-Given 81.3%
Qwen3-VL-235B-A22B Direct 80.9%
Qwen3-VL-8B-Instruct Rule-Given 80.2%
Gemma 3 12B Rule-Given 76%
Qwen2.5-VL-7B Rule-Given 73.1%
InternVL3-8B Direct 70%
Llama 3.2 11B Rule-Given 68.9%
InternVL3-8B Reasoning 66.6%
Qwen2.5-VL-7B Direct 65.3%
Gemma 3 12B Reasoning 65.3%
Qwen3-VL-8B-Instruct Direct 64.8%
Gemma 3 12B Direct 64.5%
llava-v1.6-mistral-7b Rule-Given 62.8%
Qwen2.5-VL-7B Reasoning 56%
InternVL3.5-8B Rule-Given 53.8%
Llama 3.2 11B Reasoning 52.8%
Llama 3.2 11B Direct 51.2%
llava-v1.6-mistral-7b Reasoning 47.3%
Qwen3-VL-8B-Instruct Reasoning 44.4%
llava-v1.6-mistral-7b Direct 41.7%
InternVL3.5-8B Direct 38%
InternVL3.5-8B Reasoning 37.7%

Number one over time

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

OpenAI
Own leaderboard Passed 85% in Jun 2026, by GPT-5.4
GPT-5.4 91.9%

Source: Official GeoDrive-Bench leaderboard

Qwen3-VL-235B-A22B on the official leaderboard passed 85% in September 2025

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% 2024 2025 2026 Near the ceiling (85%) GPT-5.4 Qwen3-VL-235B-A22B InternVL3-8B Gemma 3 12B Llama 3.2 11B llava-v1.6-mistral-7b 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) GPT-5.4 InternVL3-8B

Source: Official GeoDrive-Bench 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.

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 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • Pass Independent runs Scores from the official leaderboard (9 entries).
  • Pass Models scored 9 models we track.
  • Note Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. That’s expected for a saturated test.
  • Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is GPT-5.4 (#28).
  • Note Headroom Saturated. The best is 91.9%, past 85% since September 2025.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 27 rows for models we track at the last read.
  • Pass Last read November 20th, 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