Tasks and inputs
Fara 1.5 4B is listed for Image-text-to-text, Computer Use. It accepts text, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Oct 8, 2026
Fara 1.5 4B is a screenshot-based browser agent checkpoint that generates coordinate-grounded actions through Microsoft’s Fara harness.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Fara 1.5 4B is a screenshot-based browser agent checkpoint that generates coordinate-grounded actions through Microsoft’s Fara harness.
Fara 1.5 reads the user’s goal, current screenshots and prior action history, then produces reasoning and an XML-tagged tool-call block. Its visual perception works from screenshots rather than requiring DOM access. The checkpoint supports a 262,144-token context.
Microsoft recommends the Fara CLI or MagenticLite, which supplies the action executor, browser environment and controls expected by the model. The checkpoint alone does not move a pointer or navigate a browser. Keep the documented system prompt and action format so predicted coordinates are interpreted correctly.
The 4B version is a candidate when GPU memory and local deployment are priorities. Its complete multimodal weight set is larger than the language-backbone label alone suggests.
Test a form draft, a multi-step information lookup and a task that encounters an unexpected dialog. Review the resulting page state, not only the model’s description of its actions. The harness should preserve the documented pauses at critical actions and expose a trace that a reviewer can inspect. The weights are MIT-licensed; browser operations and GPU serving costs depend on your deployment.
Compare the model on the work you need to complete.
Fara 1.5 4B is listed for Image-text-to-text, Computer Use. It accepts text, image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 262.14K context. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under MIT. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Microsoft Research
Fara 1.5 27B is a screenshot-based browser agent checkpoint that generates coordinate-grounded actions through Microsoft’s Fara harness.