# Fara-7B: An Efficient Agentic Model for Computer Use Fara-7B Performance [![Microsoft](https://img.shields.io/badge/Microsoft-Project-0078D4?logo=microsoft)](https://aka.ms/msaif/fara) [![Hugging Face Model](https://img.shields.io/badge/🤗-Model-yellow)](https://huggingface.co/microsoft/Fara-7b) [![Foundry](https://img.shields.io/badge/Azure-Foundry-0089D6)](https://aka.ms/foundry-fara-7b) [![Dataset](https://img.shields.io/badge/🤗-WebTailBench%20Dataset-orange)](https://huggingface.co/datasets/microsoft/WebTailBench)
--- ## Overview **Fara-7B** is Microsoft's first **agentic small language model (SLM)** designed specifically for computer use. With only 7 billion parameters, Fara-7B is an ultra-compact Computer Use Agent (CUA) that achieves state-of-the-art performance within its size class and is competitive with larger, more resource-intensive agentic systems. Try Fara-7B locally as follows (see [Installation](##Installation) for detailed instructions): ```bash # 1. Clone repository git clone https://github.com/microsoft/fara.git cd fara # 2. Setup environment python3 -m venv .venv source .venv/bin/activate pip install -e . playwright install ``` Then in one process, host the model: ```bash vllm serve "microsoft/Fara-7B" --port 5000 --dtype auto ``` Then you can iterative query it with: ```bash fara-cli --task "whats the weather in new york now" ``` Hint: might need to do `--tensor-parallel-size 2` with vllm command if you run out of memory ### What Makes Fara-7B Unique Unlike traditional chat models that generate text-based responses, Fara-7B leverages computer interfaces—mouse and keyboard—to perform multi-step tasks on behalf of users. The model: - **Operates visually** by perceiving webpages and taking actions like scrolling, typing, and clicking on directly predicted coordinates - **Uses the same modalities as humans** to interact with computers—no accessibility trees or separate parsing models required - **Enables on-device deployment** due to its compact 7B parameter size, resulting in reduced latency and improved privacy as user data remains local - **Completes tasks efficiently**, averaging only ~16 steps per task compared to ~41 for comparable models Fara-7B is trained using a novel synthetic data generation pipeline built on the [Magentic-One](https://www.microsoft.com/en-us/research/articles/magentic-one-a-generalist-multi-agent-system-for-solving-complex-tasks/) multi-agent framework, with 145K trajectories covering diverse websites, task types, and difficulty levels. The model is based on [Qwen2.5-VL-7B](https://arxiv.org/abs/2502.13923) and trained with supervised fine-tuning. ### Key Capabilities Fara-7B can automate everyday web tasks including: - Searching for information and summarizing results - Filling out forms and managing accounts - Booking travel, movie tickets, and restaurant reservations - Shopping and comparing prices across retailers - Finding job postings and real estate listings ### Performance Highlights Fara-7B achieves state-of-the-art results across multiple web agent benchmarks, outperforming both comparable-sized models and larger systems: | Model | Params | WebVoyager | Online-M2W | DeepShop | WebTailBench | |-------|--------|------------|------------|----------|--------------| | **SoM Agents** | | | | | | | SoM Agent (GPT-4o-0513) | - | 90.6 | 57.7 | 49.1 | 60.4 | | SoM Agent (o3-mini) | - | 79.3 | 55.4 | 49.7 | 52.7 | | SoM Agent (GPT-4o) | - | 65.1 | 34.6 | 16.0 | 30.8 | | GLM-4.1V-9B-Thinking | 9B | 66.8 | 33.9 | 32.0 | 22.4 | | **Computer Use Models** | | | | | | | OpenAI computer-use-preview | - | 70.9 | 42.9 | 24.7 | 25.7 | | UI-TARS-1.5-7B | 7B | 66.4 | 31.3 | 11.6 | 19.5 | | **Fara-7B** | **7B** | **73.5** | **34.1** | **26.2** | **38.4** | *Table: Online agent evaluation results showing success rates (%) across four web benchmarks. Results are averaged over 3 runs.* ### WebTailBench: A New Benchmark for Real-World Web Tasks We are releasing **[WebTailBench](https://huggingface.co/datasets/microsoft/WebTailBench)**, a new evaluation benchmark focusing on 11 real-world task types that are underrepresented or missing in existing benchmarks. The benchmark includes 609 tasks across diverse categories, with the first 8 segments testing single skills or objectives (usually on a single website), and the remaining 3 evaluating more difficult multi-step or cross-site tasks. #### WebTailBench Detailed Results | Task Segment | Tasks | SoM GPT-4o-0513 | SoM o3-mini | SoM GPT-4o | GLM-4.1V-9B | OAI Comp-Use | UI-TARS-1.5 | **Fara-7B** | |--------------|-------|-----------------|-------------|------------|-------------|--------------|-------------|-------------| | **Single-Site Tasks** | | Shopping | 56 | 62.5 | 71.4 | 38.1 | 31.0 | 42.3 | 41.1 | **52.4** | | Flights | 51 | 60.1 | 39.2 | 11.1 | 10.5 | 17.6 | 10.5 | **37.9** | | Hotels | 52 | 68.6 | 56.4 | 31.4 | 19.9 | 26.9 | 35.3 | **53.8** | | Restaurants | 52 | 67.9 | 59.6 | 47.4 | 32.1 | 35.9 | 22.4 | **47.4** | | Activities | 80 | 70.4 | 62.9 | 41.7 | 26.3 | 30.4 | 9.6 | **36.3** | | Ticketing | 57 | 58.5 | 56.7 | 37.4 | 35.7 | 49.7 | 30.4 | **38.6** | | Real Estate | 48 | 34.0 | 17.4 | 20.1 | 16.0 | 9.0 | 9.7 | **23.6** | | Jobs/Careers | 50 | 49.3 | 44.0 | 32.7 | 22.7 | 20.7 | 20.7 | **28.0** | | **Multi-Step Tasks** | | Shopping List (2 items) | 51 | 66.0 | 62.7 | 17.0 | 7.8 | 34.0 | 20.9 | **49.0** | | Comparison Shopping | 57 | 67.3 | 59.1 | 27.5 | 22.8 | 1.2 | 8.8 | **32.7** | | Compositional Tasks | 55 | 51.5 | 39.4 | 26.7 | 17.0 | 10.3 | 9.1 | **23.0** | | **Overall** | | Macro Average | 609 | 59.7 | 51.7 | 30.1 | 22.0 | 25.3 | 19.9 | **38.4** | | Micro Average | 609 | 60.4 | 52.7 | 30.8 | 22.4 | 25.7 | 19.5 | **38.4** | *Table: Breakdown of WebTailBench results across all 11 segments. Success rates (%) are averaged over 3 independent runs. Fara-7B achieves the highest performance among computer-use models across all task categories.* **Coming Soon:** - Task Verification pipeline for LLM-as-a-judge evaluation - Official human annotations of WebTailBench (in partnership with BrowserBase) ### Evaluation Infrastructure Our evaluation setup leverages: 1. **Playwright** - A cross-browser automation framework that replicates browser environments 2. **Abstract Web Agent Interface** - Allows integration of any model from any source into the evaluation environment 3. **Fara-Agent Class** - Reference implementation for running the Fara model > **Note:** Fara-7B is an experimental release designed to invite hands-on exploration and feedback from the community. We recommend running it in a sandboxed environment, monitoring its execution, and avoiding sensitive data or high-risk domains. --- ## Installation Install the package using either UV or pip: ```bash uv sync --all-extras ``` or ```bash pip install -e . ``` Then install Playwright browsers: ```bash playwright install ``` --- ## Hosting the Model **Recommended:** The easiest way to get started is using Azure Foundry hosting, which requires no GPU hardware or model downloads. Alternatively, you can self-host with VLLM if you have GPU resources available. ### Azure Foundry Hosting (Recommended) Deploy Fara-7B on [Azure Foundry](https://ai.azure.com/explore/models/Fara-7B/version/2/registry/azureml-msr) without needing to download weights or manage GPU infrastructure. **Setup:** 1. Deploy the Fara-7B model on Azure Foundry and obtain your endpoint URL and API key 2. Add your endpoint details to the existing `endpoint_configs/` directory (example configs are already provided): ```bash # Edit one of the existing config files or create a new one # endpoint_configs/fara-7b-hosting-ansrz.json (example format): { "model": "Fara-7B", "base_url": "https://your-endpoint.inference.ml.azure.com/", "api_key": "YOUR_API_KEY_HERE" } ``` 3. Run the Fara agent: ```bash fara-cli --task "how many pages does wikipedia have" --start_page "https://www.bing.com" ``` That's it! No GPU or model downloads required. ### Self-hosting with VLLM If you have access to GPU resources, you can self-host Fara-7B using VLLM. This requires a GPU machine with sufficient VRAM. All that is required is to run the following command to start the VLLM server: ```bash vllm serve "microsoft/Fara-7B" --port 5000 --dtype auto ``` ### Testing the Fara Agent Run the test script to see Fara in action: ```bash fara-cli --task "how many pages does wikipedia have" --start_page "https://www.bing.com" --endpoint_config endpoint_configs/azure_foundry_config.json [--headful] [--downloads_folder "/path/to/downloads"] [--save_screenshots] [--max_rounds 100] [--browserbase] ``` In self-hosting scenario the `endpoint_config` points to `endpoint_configs/vllm_config.json` from the VLLM server above. If you set `--browserbase`, export environment variables for the API key and project ID. #### Expected Output ``` Initializing Browser... Browser Running... Starting Fara Agent... ########################################## Task: how many pages does wikipedia have ########################################## Running Fara... Thought #1: To find the current number of Wikipedia pages, I'll search for the latest Wikipedia page count statistics. Action #1: executing tool 'web_search' with arguments {"action": "web_search", "query": "Wikipedia total number of articles"} Observation#1: I typed 'Wikipedia total number of articles' into the browser search bar. Thought #2: Wikipedia currently has 7,095,446 articles. Action #2: executing tool 'terminate' with arguments {"action": "terminate", "status": "success"} Observation#2: Wikipedia currently has 7,095,446 articles. Final Answer: Wikipedia currently has 7,095,446 articles. Enter another task (or press Enter to exit): ``` --- # Reproducibility We provide a framework in `webeval/` to reproduce our results on WebVoyager and OnlineMind2Web. Agentic evaluations on live websites present unique challenges due to day-to-day changes. We implement several measures to ensure reliable and comparable evaluations: **BrowserBase Integration** We employ BrowserBase to manage browser session hosting, enabling reliable browser instance management. **Time-sensitive Task Updates** Tasks in benchmarks like WebVoyager can become stale or impossible. We: - Removed ~48 impossible tasks from the original WebVoyager benchmark - Updated ~50 tasks with future dates to keep them achievable - Example: *"Search for a hotel in Bali from Jan 1 to Jan 4, 2024"* → *"Search for a hotel in Bali from Jan 1 to Jan 4, 2026"* - Our updated WebVoyager benchmark is available at `webeval/data/webvoyager/WebVoyager_data_08312025.jsonl` **Environment Error Handling** Browser errors (connection drops, page timeouts) are handled robustly: - Trajectories are retried up to 5 times when environment errors occur - Complete yet incorrect trajectories are never retried - Each retry starts with a fresh browser session, with no retained state **Step Budget** Each trajectory is capped at a maximum of 100 actions across all online benchmarks. Trajectories exceeding this budget without choosing to stop are considered incorrect. ## WebEval Package Installation ```bash conda create --name fara_webeval python=3.12 conda activate fara_webeval # Install fara package pip install -e . # Install autogen submodule git submodule update --init --recursive cd autogen/python/packages pip install -e autogen-core pip install -e autogen-ext # Install webeval cd webeval pip install -e . # Install playwright playwright install ``` ## Running Evaluations Navigate to the scripts directory: ```bash cd webeval/scripts ``` Make sure you set a valid OpenAI GPT-4o endpoint in `endpoint_configs_gpt4o/dev` in order to run the WebVoyager LLM-as-a-judge! **Option 1: Self-hosted VLLM** ```bash python webvoyager.py --model_url /path/where/you/want/to/download/model/ --model_port 5000 --eval_oai_config ../endpoint_configs_gpt4o/dev/ --out_url /path/to/save/eval/files --device_id 0,1 --processes 1 --run_id 1 --max_rounds 100 ``` **Option 2: Azure Foundry Deployment** Deploy [Fara-7B on Foundry endpoint(s)](https://ai.azure.com/explore/models/Fara-7B/version/2/registry/azureml-msr), then place endpoint URLs and keys in JSONs under `endpoint_configs/`: ```bash python webvoyager.py --model_endpoint ../../endpoint_configs/ --eval_oai_config ../endpoint_configs_gpt4o/dev/ --out_url /path/to/save/eval/files --processes 1 --run_id 1_endpoint --max_rounds 100 ``` ### Notes - We use the same LLM-as-a-judge prompts and model (GPT-4o) as WebVoyager, hence the `--eval_oai_config` argument - Set `--browserbase` for browser session management (requires exported API key and project ID environment variables) - Avoid overloading a single VLLM deployment with more than ~10 concurrent processes due to known issues - See debugging output in `fara/webeval/scripts/stdout.txt` --- ## Analyzing Evaluation Results ### Evaluation Output Structure Evaluation results are stored under `--out_url` in folders organized by: - Model name - Dataset - Username - Run ID Example path: ``` /runs/WebSurfer-fara-100-max_n_images-3/fara-7b//WebVoyager_WebVoyager_data_08312025.jsonl/ ``` Each evaluation folder contains: - `gpt_eval/` - LLM-as-a-judge evaluation results - `traj/` - Per-task trajectory subdirectories containing: - `final_answer.json` (e.g., `Amazon--1_final_answer.json`) - `` indicates abortion or step budget exceeded - `scores/gpt_eval.json` - LLM judge scores - `web_surfer.log` - Action history and errors - `screenshot_X.png` - Screenshots captured before each action X ### Running Analysis Use the analysis notebook to compute metrics: ```bash cd webeval/scripts/analyze_eval_results/ jupyter notebook analyze.ipynb ``` The script: - Identifies trajectories aborted mid-execution and diagnostic reasons - Computes average scores across non-aborted trajectories - Distinguishes between aborted trajectories (errors during sampling) and completed trajectories (with terminate() call or step budget exceeded) To re-run failed tasks, execute the evaluation script again with the same `run_id` and `username` - it will skip non-aborted tasks.
Example WebVoyager GPT Eval Result ```json { "score": 1.0, "gpt_response_text": "To evaluate the task, we need to verify if the criteria have been met:\n\n1. **Recipe Requirement**: A vegetarian lasagna recipe with zucchini and at least a four-star rating.\n\n2. **Search and Results**:\n - The screenshots show that the search term used was \"vegetarian lasagna zucchini.\"\n - Among the search results, \"Debbie's Vegetable Lasagna\" is prominently featured.\n \n3. **Evaluation of the Recipe**:\n - Rating: \"Debbie's Vegetable Lasagna\" has a rating of 4.7, which satisfies the requirement of being at least four stars.\n - The presence of zucchini in the recipe is implied through the search conducted, though the screenshots do not explicitly show the ingredients list. However, the result response confirms the match to the criteria.\n\nGiven the information provided, the task seems to have fulfilled the requirement of finding a vegetarian lasagna recipe with zucchini and a four-star rating or higher. \n\n**Verdict: SUCCESS**" } ```
--- ## Citation If you use Fara in your research, please cite our work: ---