Primary use stated or inferred cautiously from official documentation.
Ai2
AI2 / allenai/Molmo2-4B-tb
Molmo2-4B-tb
ToyBlocks high-level planner derived from Molmo2-4B and fine-tuned for block stacking. Ai2 released the weights on November 5. The roughly 4.5-billion-parameter checkpoint reads the goal and front camera and predicts the next step and target point.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | allenai/Molmo2-4B-tb |
|---|---|
| Model type | Robotics · Multimodal |
| Access model | Open weights |
| License | Apache 2.0 |
| Deployment | Self-hosted |
| Release | 05 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Image |
| Exit | Plan, discrete action and target point |
| Context window | One 640 × 360 front image and a text goal |
| maximum output | Not published |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · PyTorch |
Study hierarchical planning and visual grounding on a tabletop. The model selects a discrete action, such as picking or rotating a block. A motion planner or policy such as ManiFlow-tb performs it. The two levels can be compared without mixing their results.
Uses the native Molmo2 format and ToyBlocks code. The card estimates about 20 GB of GPU memory in BF16. It is limited to tabletop blocks; it does not send motor commands or demonstrate safety for other robots. Check the weight license and base-model data terms.
03 / CAPABILITIES
What it can do
Orientation
Study hierarchical planning and visual grounding on a tabletop. The model selects a discrete action, such as picking or rotating a block. A motion planner or policy such as ManiFlow-tb performs it. The two levels can be compared without mixing their results.
Context
One 640 × 360 front image and a text goal · Not published
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Input modalities
TextImage
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
No hosted commercial API has been verified for this checkpoint. Downloadable weights do not cover storage, accelerators, memory or operating costs.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
On easy simulations with seen blocks, Ai2 reports 97.5% with a motion planner and 55.5% with ManiFlow-tb over 200 episodes. The physical robot with ManiFlow-tb succeeded in 6 of 10 trials. These are full systems with different executors and do not measure the planner alone.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| ToyBlocks Easy seen (motion planner) | 97,5 % | Score | View source ↗ |
| ToyBlocks Easy seen (ManiFlow) | 55,5 % | Score | View source ↗ |
| ToyBlocks real robot (Molmo2 + ManiFlow) | 6 / 10 | Score | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Molmo2-4B-tb
Version added to Inferama's verified catalog.
07 / SOURCES