Visual Identification Benchmark
The Identification task asks each model to recognize and name a specific entity in an image based on visual evidence. The answer might be a brand, an object type, a color, a material, or a printed label. The model must single out the correct target among look-alikes and distractors and return the requested name or descriptor.
Evals updated August 14, 2026Pricing updated August 18, 2026
| 1 | 100.0% | 1.4K | $0.0042 | 2.71s | ||
| 1 | 100.0% | 1.3K | $0.014 | 5.60s | ||
| 1 | 100.0% | 1.5K | $0.0070 | 5.93s | ||
| 4 | 96.9% | 1.3K | $0.0015 | 2.28s | ||
| 4 | 96.9% | 1.3K | $0.0007 | 18.17s | ||
| 6 | 93.8% | 1.2K | $0.0009 | 2.23s | ||
| 6 | 93.8% | 395 | $0.0012 | 2.81s | ||
| 8 | 90.6% | 976 | $0.0002 | 3.23s | ||
| 8 | 90.6% | 1.3K | $0.0085 | 4.15s | ||
| 8 | 90.6% | 1.8K | $0.0038 | 5.34s | ||
| 8 | 90.6% | 1.1K | $0.0027 | 7.16s | ||
| 12 | 87.5% | 1.7K | $0.0032 | 6.96s | ||
| 13 | Qwen 3.7 Plus | 84.4% | 981 | $0.0003 | 2.21s | |
| 13 | 84.4% | 981 | <$0.0001 | 3.54s | ||
| 13 | 84.4% | 1.3K | $0.0072 | 3.71s | ||
| 13 | GLM 5V Turbo | 84.4% | 1.3K | $0.0015 | 4.59s | |
| 17 | 81.3% | 1.2K | $0.0004 | 1.23s | ||
| 17 | 81.3% | 1.4K | $0.0006 | 1.87s | ||
| 17 | 81.3% | 1.3K | $0.0027 | 2.33s | ||
| 17 | 81.3% | 1.2K | $0.0035 | 3.35s | ||
| 17 | 81.3% | 1.3K | $0.0041 | 5.21s | ||
| 22 | Qwen3.8 27BNEW | 78.1% | 981 | $0.0005 | 1.13s | |
| 22 | 78.1% | 978 | $0.0002 | 1.97s | ||
| 22 | 78.1% | 1.3K | $0.0013 | 2.73s | ||
| 22 | 78.1% | 1.3K | $0.0004 | 3.43s | ||
| 22 | 78.1% | 1.3K | $0.0039 | 3.52s | ||
| 22 | Grok 4.6NEW | 78.1% | 1.6K | $0.0048 | 5.33s | |
| 22 | 78.1% | 1.7K | $0.0045 | 6.84s | ||
| 29 | 75.0% | 1.3K | $0.0067 | 2.30s | ||
| 30 | Kimi K2.6 | 62.5% | 1.3K | $0.0010 | 5.16s |
Score vs. cost
Identification score (Accuracy) against estimated cost per sample. Upper-left is the sweet spot: high quality at low cost.
30 models on the current benchmark · Identification task only
Example Identification benchmark tasks
Real samples from the benchmark: the image each model sees, the question it is asked, and the ground-truth answer it is scored against.
The models are asked
What brand name is printed on the yogurt labels? Output the exact text exactly as printed.
Ground truth
Yoplait
The models are asked
What color is the vehicle parked immediately to the left of the bright green car? Answer strictly with the lowercase color name.
Ground truth
black
The models are asked
What specific type of red fruit is present in both the loose rectangular metal container on the left and inside the small white portion bowls? Answer strictly with the lowercase singular name of the fruit.
Ground truth
strawberry
How Identification is scored
Each answer is graded against the ground-truth name or descriptor. The leaderboard score is the percentage of samples identified correctly.
Every model runs the same sample set in a single evaluation pass. Token usage is measured from each provider’s API response, and cost per sample is that usage multiplied by the model’s published pricing. See the full methodology.
Frequently Asked Questions
Each model answers the same identification prompts. Answers are graded against the ground-truth name or descriptor, and the score is the percentage answered correctly.
Identification asks what something is: a name, brand, color, or material. Detection asks what and where, requiring a bounding box for every object. A model can be good at naming things while being poor at localizing them.
Most models in this leaderboard link to their Playground page. Click the model name to open it, then upload your own image and run it. A few models are benchmarked for comparison only and do not have a Playground page yet.