Most internet brokers at present drive a browser one motion at a time. The mannequin receives the present web page state — as a screenshot or DOM textual content — and predicts the subsequent click on, keypress, or scroll. This action-at-a-time design made sense when language fashions had restricted reasoning capability. As fashions have turn into extra succesful at writing and debugging code, that inflexible loop has turn into a constraint slightly than a construction that helps.
Microsoft Analysis’s AI Frontiers lab constructed a unique strategy. Their new open-source framework, Webwright, provides the agent a terminal as a substitute of a stateful browser session. The agent writes Playwright code to regulate browsers, runs bash instructions, inspects logs, and iteratively refines scripts. Playwright is an open-source browser automation library, additionally from Microsoft, that helps programmatic management of Chromium, Firefox, and WebKit browsers.
What Webwright Does In a different way
Webwright separates the agent from the browser and treats the browser as one thing the agent can launch, examine, and discard whereas growing a program. The persistent artifact will not be the browser session however the code and logs within the native workspace.
This is similar mannequin a developer makes use of when writing an RPA (Robotic Course of Automation) script. As an alternative of manually clicking by a website every time, they write a script as soon as. That script could be rerun, tailored, and shared. Webwright applies this to LLM-powered brokers.
The system has three core parts: a Runner, a Mannequin Endpoint, and a terminal Setting. The runner is about 150 strains of code, the mannequin interface about 550 strains, and the surroundings about 300 strains. There is no such thing as a multi-agent orchestration or advanced planning hierarchy — only a single agent loop.
All intermediate code, logs, screenshots, and outcomes are saved within the workspace, making every run simple to examine.
The Agent Loop
The Runner sends the present context to the mannequin. The mannequin returns a considering block and a shell command. That command runs within the Setting, which returns terminal output, logs, screenshots, or error tracebacks. These observations return into context, and the loop continues.
Moderately than issuing one primitive motion at a time, a coding agent can naturally specific multi-step interactions — resembling deciding on a date or filling out a whole type — as a compact program. Loops, features, and abstractions permit the agent to generalize throughout comparable duties with out repeatedly predicting comparable sequences of low-level steps.
Two Engineering Challenges
Untimely ‘finished’ and context explosion are the 2 core points. With open-ended bash actions, the mannequin should self-report completion and infrequently claims success with out really ending. They added a gate: the agent should generate a self-reflection config, run a ultimate script in a contemporary folder with logs and screenshots, and cross its personal self-reflection judgement that outputs success or failure earlier than emitting finished: true. In any other case, the flag is dropped and it retries.
For context size, lengthy coding trajectories rapidly exceed context limits, so that they compact historical past each 20 steps right into a single abstract.
Benchmark Outcomes
Webwright was evaluated on two benchmarks: On-line-Mind2Web and Odysseys.
On-line-Mind2Web comprises 300 duties throughout 136 broadly used websites and makes use of an automatic LLM-as-a-Decide analysis framework. GPT-5.4 achieves 86.67% general accuracy, representing the best amongst all open-sourced harness recipes within the AutoEval class of the On-line-Mind2Web benchmark, with a 100-step finances. Claude Opus 4.7 reached 84.7% general however carried out higher on arduous duties at N=100 steps — 80.5% versus 76.6% for GPT-5.4.
In addition they reproduced a GPT-5.4 baseline in a traditional screenshot-based agent setting, the place the mannequin predicts x,y coordinates for clicks and typing actions. Utilizing the identical underlying mannequin, Webwright achieves substantial features throughout all three problem classes, highlighting the good thing about the code-driven terminal-based strategy over step-by-step coordinate prediction.
Odysseys evaluates long-horizon shopping duties spanning a number of web sites. Duties common 272.3 phrases of directions. Within the April 2026 leaderboard, the best-performing mannequin was Opus 4.6, with a high rating of 44.5. Webwright powered by GPT-5.4 reaches 60.1%, a 35.1% relative enchancment over the earlier state-of-the-art. In comparison with the bottom GPT-5.4 efficiency of 33.5%, this corresponds to a 79.4% relative enchancment — or 26.6 absolute factors.
Price Evaluation
Claude Opus 4.7 is extra environment friendly within the variety of steps to unravel every job (imply 21.9 steps) in comparison with GPT-5.4 (imply 26.3 steps). Nonetheless, Claude Opus 4.7 is priced considerably greater in comparison with GPT-5.4 ($5 vs. $2.50 per 1M enter tokens, and $25 vs. $15.00 per 1M output tokens, April 2026), which makes the typical per-task value greater in comparison with GPT-5.4 ($2.37 vs. $6.09). The primary 50 steps ship 82% accuracy, and the subsequent 50 steps ship 3–4 extra factors.
Small Mannequin Efficiency
The analysis group additionally examined Qwen3.5-9B on the arduous cut up of On-line-Mind2Web. When duties are augmented with pre-built reusable device scripts, Qwen3.5-9B achieves 66.2% on On-line-Mind2Web web sites with greater than 5 instruments. This exhibits that smaller, lower-cost fashions can deal with advanced internet duties when paired with a pre-built device library.
Marktechpost’s Visible Explainer
Webwright
Fast Begin Information
What Is Webwright?
Webwright is an open-source, terminal-native internet agent framework from Microsoft Analysis. As an alternative of predicting one browser click on at a time, the agent writes Playwright code, runs bash instructions, and shops reusable scripts in a neighborhood workspace.
- ~1,000 strains of harness code throughout 3 modules — no hidden orchestration
- Single agent loop: Runner, Mannequin Endpoint, and terminal Setting
- 86.7% on On-line-Mind2Web | 60.1% on Odysseys with GPT-5.4
- Backends: OpenAI, Anthropic, OpenRouter
- Scripts reusable in Claude Code, Codex, OpenClaw
# GitHub repository
github.com/microsoft/Webwright
What You Want Earlier than Putting in
Verify the next are prepared earlier than working any set up instructions.
- Python 3.10+ — required minimal runtime
- Chromium — put in through Playwright within the subsequent step
- API key — OpenAI, Anthropic, or OpenRouter
- Git — to clone the repository
# Verify your Python model python --version # Should return Python 3.10 or greater
Clone and Set up Webwright
Clone the repo, set up in editable mode, then set up Chromium for Playwright browser management.
# 1. Clone the repository git clone https://github.com/microsoft/Webwright cd Webwright # 2. Set up the package deal in editable mode pip set up -e . # 3. Set up Chromium for Playwright playwright set up chromium
The -e flag means native supply edits apply instantly with out reinstalling.
Run Your First Internet Process
Export your API key, then cross a job instruction and begin URL to the CLI.
# Export your key export OPENAI_API_KEY="sk-..." export ANTHROPIC_API_KEY="sk-ant-..." # Run a job python -m webwright.run.cli -c base.yaml -c model_openai.yaml -t "Discover most cost-effective financial system flight SEA to JFK on 2026-05-15" --start-url https://www.google.com/flights --task-id demo_openai -o outputs/default
| Flag | Description |
|---|---|
| -c | Config file from src/webwright/config/ — stackable |
| -t | Process instruction in plain English |
| –start-url | Preliminary URL for the browser session |
| –task-id | Output subfolder identify |
| -o | Root output listing for logs and scripts |
Use Webwright as a Claude Code Ability
Webwright ships a built-in Claude Code ability. No separate LLM API secret’s wanted past your Claude Code subscription. Claude Code reads PNG screenshots natively.
# Undertaking-scoped (inside this repo solely) mkdir -p .claude/expertise .claude/instructions ln -s "$PWD/expertise/webwright" .claude/expertise/webwright ln -s "$PWD/expertise/webwright/instructions" .claude/instructions/webwright # Person-scoped (all initiatives) mkdir -p ~/.claude/expertise ~/.claude/instructions ln -s "$PWD/expertise/webwright" ~/.claude/expertise/webwright ln -s "$PWD/expertise/webwright/instructions" ~/.claude/instructions/webwright
Restart Claude Code after putting in, then use slash instructions:
# One-shot job /webwright:run search Google Flights SEA to JFK 2026-05-15 # Reusable parameterized CLI device /webwright:craft search a ticket from LAX to SFO depart June 7
Key Takeaways
- Webwright makes use of a terminal loop the place the agent writes and runs Playwright code as a substitute of predicting one browser motion at a time.
- GPT-5.4 reached 86.7% on On-line-Mind2Web (100-step finances) and 60.1% on Odysseys — 26.6 factors above the bottom GPT-5.4 rating of 33.5%.
- The harness is ~1,000 strains throughout three modules with no multi-agent orchestration.
- Qwen3.5-9B reached 66.2% on the arduous cut up of On-line-Mind2Web when augmented with pre-built device scripts.
- Process scripts are packaged as reusable CLIs, shareable throughout Claude Code, Codex, and OpenClaw.
Try the Repo and Technical particulars. Additionally, be happy to observe us on Twitter and don’t overlook to hitch our 150k+ ML SubReddit and Subscribe to our Publication. Wait! are you on telegram? now you’ll be able to be part of us on telegram as nicely.
Have to companion with us for selling your GitHub Repo OR Hugging Face Web page OR Product Launch OR Webinar and so on.? Join with us
