Skip to main content
BenchLM

CountBench

We show this table for reference; we do not rank on it.

Data verified 40 confirmed releases in the last 30 daysFollow model changes

A visual counting benchmark that tests whether a model can count objects and entities reliably in complex scenes.

Benchmark score on CountBench — September 22, 2026

We compile the CountBench rows from provider self-reports. Qwen3.6-27B leads the table at 97.8%, followed by ZAYA1-VL-8B (88.1%) and LFM2.5-VL-3B (87.3%). We do not use these results to rank models overall.

1Open weights

Qwen3.6-27B

Alibaba

97.8%
Overall 46.27Context 262K
2Open weights

ZAYA1-VL-8B

Zyphra

88.1%
Context 131K
3Open weights

LFM2.5-VL-3B

LiquidAI

87.3%
Context 32K

4 modelsMultimodal & GroundedCurrentDisplay onlyUpdated September 22, 2026

Benchmark score table (4 models)

Score
1
Qwen3.6-27BAlibaba · Open weight
97.8%
2
ZAYA1-VL-8BZyphra · Open weight
88.1%
3
LFM2.5-VL-3BLiquidAI · Open weight
87.3%
4
LFM2.5-VL-450MLiquidAI · Open weight
73.3%

Among the reported CountBench rows, Qwen3.6-27B is first at 97.8%. The third row is 10.5 points behind. The broader top-10 range is 24.5 points, so the table still separates the published systems.

4 models have been evaluated on CountBench. The benchmark falls in the Multimodal & Grounded category. CountBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About CountBench

Year

2026

Tasks

Visual counting tasks

Format

Image-grounded counting

Difficulty

Fine-grained visual perception

Counting failures are a common multimodal weakness even in otherwise strong models. CountBench isolates that skill and makes it easy to compare raw perception accuracy across models.

Qwen3.6 launch benchmarks

Freshness and provenance

Version

CountBench 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does CountBench measure?

A visual counting benchmark that tests whether a model can count objects and entities reliably in complex scenes.

Which model scores highest on CountBench?

Qwen3.6-27B by Alibaba currently leads with a score of 97.8% on CountBench.

How many models are evaluated on CountBench?

4 AI models have been evaluated on CountBench on BenchLM.

Compare top models on CountBench

Qwen3.6-27B vs ZAYA1-VL-8BZAYA1-VL-8B vs LFM2.5-VL-3BLFM2.5-VL-3B vs LFM2.5-VL-450M
Last updated: September 22, 2026 · BenchLM version CountBench 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.