Specialist modelActive
Perceptron Mk1.5
Perceptron
- Released
- September 25, 2026
- Data date
- October 3, 2026
- Model class
Perceptron Mk1.5 is a multimodal model for embodied agents. It accepts text, images, video, and audio, then emits text, points, boxes, polygons, clips, and time-based object tracks. Perceptron names drones, robot dogs, smart glasses, and smartphones as deployment environments.
Access is provided through the Perceptron Platform and SDK under the API model ID perceptron-mk1.5. Perceptron specifies 32K tokens of multimodal context. The model understands audio inputs and supports tool calls. API pricing is $0.15 per 1M input tokens and $1.50 per 1M output tokens.
Specifications and access
| Specification | Value and source |
|---|---|
| Model class | Multimodal control model for embodied agentsSource |
| Modalities | Text, images, video, and audio as inputSource |
| Outputs | Text, points, boxes, polygons, clips, and object tracksSource |
| API model ID | perceptron-mk1.5Source |
| Context window | 32K tokens of multimodal context (provider specification)Source |
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
DailyOmni accuracy (Show measurement, test conditions, and source)
- Source value
- 74.67
- Score
- 74.67
- Metric
- percent
- Unit
- %
- Category
- audio-visual
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Figure 4 audio-visual evaluation.
WorldSense accuracy (Show measurement, test conditions, and source)
- Source value
- 50.32
- Score
- 50.32
- Metric
- percent
- Unit
- %
- Category
- audio-visual
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Figure 4 audio-visual evaluation.
OmniBench accuracy (Show measurement, test conditions, and source)
- Source value
- 51.05
- Score
- 51.05
- Metric
- percent
- Unit
- %
- Category
- audio-visual
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Figure 4 audio-visual evaluation.
AVHBench accuracy (Show measurement, test conditions, and source)
- Source value
- 80.56
- Score
- 80.56
- Metric
- percent
- Unit
- %
- Category
- audio-visual
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Figure 4 audio-visual evaluation.
LiveVQA-W accuracy (Show measurement, test conditions, and source)
- Source value
- 56
- Score
- 56
- Metric
- percent
- Unit
- %
- Category
- audio-visual
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Figure 5; four samples, majority vote, and tools; mean of two benchmark levels.
EgoSchema hard accuracy, Figure 3 (Show measurement, test conditions, and source)
- Source value
- 63.75
- Score
- 63.75
- Metric
- percent
- Unit
- %
- Category
- egocentric-video
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- disputed
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Figure 3 egocentric evaluation.
- Context
- The same page's final video table reports 61.3 for EgoSchema hard. Both observations are retained because their scope is unresolved.
EgoSchema hard accuracy, final table (Show measurement, test conditions, and source)
- Source value
- 61.3
- Score
- 61.3
- Metric
- percent
- Unit
- %
- Category
- egocentric-video
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- disputed
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Final video benchmark table.
- Context
- The same page's Figure 3 reports 63.75 for EgoSchema hard. Both observations are retained because their scope is unresolved.
Molmo2-Track center-F1 (Show measurement, test conditions, and source)
- Source value
- 0.647
- Score
- 0.647
- Metric
- ratio
- Unit
- ratio
- Category
- tracking
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Figure 2 tracking evaluation.
Molmo2-Track point-HOTA (Show measurement, test conditions, and source)
- Source value
- 0.628
- Score
- 0.628
- Metric
- ratio
- Unit
- ratio
- Category
- tracking
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Perceptron
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Figure 2 tracking evaluation.
Sources and data date
Every statement links to its underlying documentation or leaderboard.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 4, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Unresolved | egoschema-hard-configuration Original wording The provider page gives 63.75 in Figure 3 and 61.3 in its final video table. Both are preserved and neither is selected as the canonical value. |
| Additional source | Perceptron Mk1.5 announcement (retrieved October 3, 2026; October 4, 2026) · Perceptron Mk1.5 announcement · Editorial description reviewed October 3, 2026 |