Skip to main content
BenchLM
Data

ERQA

Glimpse leads the ERQA leaderboard on BenchLM's November 2026 update with 83.8%, ahead of Qwen3.8 Max (77.8%) and Qwen3.8-Flash-Next (72.3%), across 19 models.

Data verified

A grounded visual reasoning benchmark focused on evidence-based question answering over real images.

Top models on ERQA — November 21, 2026

As of November 21, 2026, Glimpse leads the ERQA leaderboard with 83.8% , followed by Qwen3.8 Max (77.8%) and Qwen3.8-Flash-Next (72.3%).

19 modelsMultimodal & Grounded3% of Multimodal reference weightCurrentUpdated November 21, 2026

Leaderboard (19 models)

None of the 19 results on this page has a documented run by the benchmark or an independent evaluator: 12 reported by the labs that built the models and 7 reported in another publisher’s document.

ERQA leaderboard
RankModel / configurationScoreParameters (B)Open / closed
1GlimpseElorianReported by Elorian
83.8%
Not reportedClosed
2Qwen3.8 MaxAlibabaReported by Alibaba
77.8%
Not reportedOpen
3Qwen3.8-Flash-NextAlibabaReported by Alibaba
72.3%
Not reportedOpen
4Seed 2.1 ProByteDanceReported by ByteDance
72.0%
Not reportedClosed
5Seed 2.1 TurboByteDanceReported by ByteDance
71.3%
Not reportedClosed
6Qwen3.8-Omni-FlashAlibabaReported by Alibaba
71.0%
Not reportedClosed
7Gemini 3 ProGoogleAnother lab’s comparison table
70.5%
Not reportedClosed
8Qwen3.7 PlusAlibabaReported by Alibaba
69.8%
Not reportedClosed
9Gemini 3.1 ProGoogleReported by Meta AI (comparison table)
69.4%
Not reportedClosed
10Qwen3.5 397BAlibabaReported by Alibaba
67.5%
Not reportedOpen
11Qwen3.8-27BAlibabaReported by Alibaba
65.5%
Not reportedOpen
12GPT-5.4OpenAIReported by Meta AI (comparison table)
65.4%
Not reportedClosed
13Muse SparkMetaReported by Meta
64.7%
Not reportedClosed
14Qwen3.6-27BAlibabaReported by Alibaba
62.5%
Not reportedOpen
15GPT-5.2OpenAIAnother lab’s comparison table
59.8%
Not reportedClosed
16Grok 4.20xAIReported by Meta AI (comparison table)
54.1%
Not reportedClosed
17Claude Opus 4.6AnthropicReported by Meta AI (comparison table)
51.6%
Not reportedClosed
18LFM3-VL-2BLiquidAIReported by LiquidAI
48.2%
Not reportedOpen
19Claude Opus 4.5AnthropicAnother lab’s comparison table
46.8%
Not reportedClosed
  • Lab-reported
  • Benchmark-run
  • Independent
What the labels mean

According to BenchLM.ai, Glimpse leads the ERQA benchmark with a score of 83.8%, followed by Qwen3.8 Max (77.8%) and Qwen3.8-Flash-Next (72.3%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.

19 models have been evaluated on ERQA. The benchmark falls in the Multimodal & Grounded category. BenchAlign v6.0 gives ERQA 3% of the Multimodal reference weight, so it moves the Multimodal leaderboard. Reference weights are relative weights in the calibrated model, not fixed shares of a score.

About ERQA

Year

2026

Tasks

Evidence-based visual QA

Format

Grounded image reasoning

Difficulty

Grounded multimodal reasoning

ERQA is useful as a grounded reasoning check because it emphasizes answer correctness tied to visual evidence rather than fluent but ungrounded descriptions.

Freshness and provenance

Version

ERQA 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

Current

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 ERQA measure?

A grounded visual reasoning benchmark focused on evidence-based question answering over real images.

Which model scores highest on ERQA?

Glimpse by Elorian currently leads with a score of 83.8% on ERQA.

How many models are evaluated on ERQA?

19 AI models have published results on ERQA in the BenchLM catalog.

Last updated: November 21, 2026 · BenchLM version ERQA 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 7,000+ subscribers.

One email each week. Unsubscribe anytime.