BAAIChina2026
RoboBrain 2.5
An AI model: software that a robot uses to perceive, plan or move. This page describes the software, not a robot.
Robotsnot stated
VersionsRoboBrain 2.5-4BRoboBrain 2.5-8B
The developer has released this model in several versions. Each figure applies to the version its source names.
Key figures as stated by the developer
3 of 6 stated · 3 gaps- Model type not stated
- Controls not stated
-
Inputs A
cameralanguage - Access B open download
- Parameters A 8,000,000,000
- Base model not stated
How to read the figures
- 50HzA
- The letter in the grey square after a figure points to the source the figure came from. Pages with several sources carry several letters.
Versions and notes in the key figures (6)
Model type
Note BAAI's wording: "a next-generation embodied AI foundation model that advances general perception, spatial reasoning, and temporal modeling".
Controls
Note README, 3D trajectory example: "Output is formatted as a list of tuples" of x and y coordinates and depth; the model returns points and traces, not robot actions.
Inputs
README examples pass an image and a text prompt ("reach for the banana on the plate"); the report describes "dense, step-aware progress prediction and execution state understanding across varying viewpoints". Images = camera, text = language.
Access
Technical report: "The code and checkpoints are available at project website"; the checkpoints are on Hugging Face.
Parameters
README model table: "4B parameter version of the RoboBrain2.5" (4B) and "8B parameter version of the RoboBrain2.5" (8B). The value is the largest released version.
Versions
- RoboBrain 2.5-4B
- 4,000,000,000
- RoboBrain 2.5-8B
- 8,000,000,000
Base model
Note The README says only "RoboBrain2.5 is also compatible with the official training code for Qwen3VL."
Full specifications
4 of 12 fields stated| Field | Value | Source and caveats |
|---|---|---|
| Input and output | ||
| Model type | not stated | Note BAAI's wording: "a next-generation embodied AI foundation model that advances general perception, spatial reasoning, and temporal modeling". B |
| Inputs | camera language | A README examples pass an image and a text prompt ("reach for the banana on the plate"); the report describes "dense, step-aware progress prediction and execution state understanding across varying viewpoints". Images = camera, text = language. |
| Controls | not stated | Note README, 3D trajectory example: "Output is formatted as a list of tuples" of x and y coordinates and depth; the model returns points and traces, not robot actions. A |
| Size and speed | ||
| Parameters | 8,000,000,000 | A README model table: "4B parameter version of the RoboBrain2.5" (4B) and "8B parameter version of the RoboBrain2.5" (8B). The value is the largest released version. Versions
|
| Where it runs and what it is built on | ||
| Base model | not stated | Note The README says only "RoboBrain2.5 is also compatible with the official training code for Qwen3VL." A |
| Openness | ||
| Access | open download | B Technical report: "The code and checkpoints are available at project website"; the checkpoints are on Hugging Face. |
| License | Apache-2.0 | C Model card of the 4B version: "license: apache-2.0". The code repository carries the "Apache License Version 2.0, January 2004" licence file. |
| The developer’s own claims | ||
| Works across robots | not stated | Note The README says "task progress prediction across different tasks, viewpoints, and embodiments", but names no robot models. A Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count. |
Not stated: Degrees of freedom controlled, Control rate, Runs where, Learns new tasks
Notes
- BAAI describes RoboBrain 2.5 as a next-generation embodied AI foundation model for general perception, spatial reasoning and temporal modeling.
- The RoboBrain 2.5 technical report was submitted to arXiv on 20 January 2026.
- The Beijing Academy of Artificial Intelligence (BAAI) was founded on the initiative of the Beijing Municipal Science and Technology Commission and the Haidian District government.
Sources
-
A
Manufacturer product page. github.
com/ · RetrievedFlagOpen/ RoboBrain2. 5 -
B
arxiv.
org/ · Retrievedabs/ 2601. 14352 -
C
huggingface.
co/ · RetrievedBAAI/ RoboBrain2. 5-4B
One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.