Skip to content
#50BMay 15, 2026Evidence A

RoboBrain 2.5

BAAI · 110B and 270B tiers

An open brain for embodied intelligence

BAAI’s May 2026 open embodied multimodal model, in 110B and 270B tiers under Apache-2.0: video 80, perception 90 and agents 72 for robot manipulation and spatial planning; documents 60 and generation 30 make it unsuitable for general business work. Composite 70.6.

TextImageVideoEmbodiedNative

Last updated Oct 21, 2026Editorial scoring

Bottom line

For embodied research and robot manipulation, RoboBrain 2.5 is one of the most complete Chinese open-weight choices: 110B/270B, Apache-2.0, video 80. Its scores are not meant for general business — documents 60 and generation 30 mean disappointment for content or document work.

Eight-dimension breakdown

editorial mapping · 0–100
Perception90Document60Chart QA55Video80Agent72Generation30Chinese80Ecosystem788D
  • RoboBrain 2.570.6
Visual perception & grounding20%
90
Document parsing & Chinese OCR18%
60
Chart & document reasoning14%
55
Video & long-horizon reasoning14%
80
GUI agents & tool use14%
72
Generation & editing6%
30
Chinese context & culture6%
80
Ecosystem, cost & availability8%
78

Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.

Public benchmarks

as reported, not converted
Benchmark / metricValueNote
参数规模110B / 270BTwo tiers
许可证Apache-2.0Commercial use
上下文128KOfficial spec

Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.

Scenario mapping

five stress tests, one scale
ScenarioScoreObservation
Embodied manipulation92 · A new capability axis here
30-min video QA80 · Good dynamic-scene understanding
Chinese receipt OCR60 · Not a document model
Mobile GUI control55 · Not GUI-specific
Chinese cultural context80 · Mostly research context

Strengths and limitations

Strengths

  • Embodied manipulation and spatial planning: a scarce combination among Chinese open weights.
  • Video 80 and perception 90 make dynamic-scene understanding solid.
  • Apache-2.0 plus 110B/270B gives robotics teams their own base.

Limitations

  • Documents 60 and generation 30: unusable for general content tasks.
  • Ecosystem 78: tooling is research-oriented, with thin production docs.
  • Self-hosting 270B is expensive, and real robots still need an action-policy layer.

Price and real cost

List price

—N/A

Open weights; self-hosted, no API fee. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥0.0 (assuming input at 25% of output price).

Use your volume

About RoboBrain 2.5

Is RoboBrain 2.5 open source?

Licence: Open weights (Apache-2.0). Weights are available, but check the exact terms and filing requirements before commercial use.

How much does RoboBrain 2.5 cost?

Open weights; self-hosted, no API fee. It ranks #50 with a composite of 70.6. Real bills depend on resolution, tiling, caching and volume discounts.

What is RoboBrain 2.5 best for?

Robot manipulation and grasping; Embodied-intelligence research; Spatial planning in simulation。

Why does RoboBrain 2.5 score 70.6?

The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Visual perception & grounding; its weakest is Generation & editing.

Personal ranking · Not an official score

Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.