Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

feat(llm): default ChatBrowserUse to bu-2-0-mini-preview

Saurav Panda9be28bcea151
diff --git a/README.md b/README.md
index 9dfdae628..31286952f 100644
--- a/README.md
+++ b/README.md
@@ -113,7 +113,7 @@ async def main():
agent = Agent(
task="Find the number of stars of the browser-use repo",
llm=ChatBrowserUse(model='openai/gpt-5.5'),
- # llm=ChatBrowserUse(model='bu-2-0'), # Browser Use's optimized model
+ # llm=ChatBrowserUse(model='bu-2-0-mini-preview'), # Browser Use's optimized model
# llm=ChatOpenAI(model='gpt-5.5'),
# llm=ChatAnthropic(model='claude-opus-4-8'), # Sonnet also works well
)
diff --git a/browser_use/llm/browser_use/chat.py b/browser_use/llm/browser_use/chat.py
index 6f0d0c4b6..25e108f8b 100644
--- a/browser_use/llm/browser_use/chat.py
+++ b/browser_use/llm/browser_use/chat.py
@@ -44,7 +44,7 @@ class ChatBrowserUse(BaseChatModel):
def __init__(
self,
- model: str = 'bu-2-0',
+ model: str = 'bu-2-0-mini-preview',
api_key: str | None = None,
base_url: str | None = None,
timeout: float = 120.0,
@@ -58,7 +58,8 @@ class ChatBrowserUse(BaseChatModel):
Args:
model: Model name to use. Options:
- - 'bu-2-0' or 'bu-latest': Default model (latest premium)
+ - 'bu-2-0-mini-preview': Default model (fast + cheap, preview)
+ - 'bu-2-0' or 'bu-latest': Premium model
- 'bu-1-0': Previous generation model
- 'bu-qa-1': Website QA model (tests a site and scores functionality/aesthetics)
- 'browser-use/bu-30b-a3b-preview': Browser Use Open Source Model
@@ -73,7 +74,7 @@ class ChatBrowserUse(BaseChatModel):
"""
# Accept 'bu-*' aliases and any provider-prefixed id; the gateway resolves the
# latter (anthropic/*, openai/*, google/*, browser-use/*), so we don't enumerate them.
- bu_aliases = ['bu-latest', 'bu-1-0', 'bu-2-0', 'bu-qa-1']
+ bu_aliases = ['bu-latest', 'bu-1-0', 'bu-2-0', 'bu-2-0-mini-preview', 'bu-qa-1']
is_valid = model in bu_aliases or '/' in model
if not is_valid:
raise ValueError(
@@ -82,7 +83,8 @@ class ChatBrowserUse(BaseChatModel):
"'openai/gpt-5.5', or 'google/gemini-3-pro'."
)
- # Normalize bu-latest to the current latest model
+ # Normalize bu-latest to the current latest model. Deliberately not the constructor
+ # default: 'latest' tracks the stable premium line, not the preview.
if model == 'bu-latest':
self.model = 'bu-2-0'
else:
diff --git a/browser_use/llm/models.py b/browser_use/llm/models.py
index 81a277731..81edf9945 100644
--- a/browser_use/llm/models.py
+++ b/browser_use/llm/models.py
@@ -8,7 +8,7 @@ Usage:
model = llm.azure_gpt_4_1_mini
model = llm.openai_gpt_4o
model = llm.google_gemini_2_5_pro
- model = llm.bu_latest # or bu_1_0, bu_2_0
+ model = llm.bu_latest # or bu_2_0_mini_preview, bu_2_0, bu_1_0
"""
import os
@@ -83,6 +83,7 @@ cerebras_gemma_4_31b: 'BaseChatModel'
bu_latest: 'BaseChatModel'
bu_1_0: 'BaseChatModel'
bu_2_0: 'BaseChatModel'
+bu_2_0_mini_preview: 'BaseChatModel'
def get_llm_by_name(model_name: str):
@@ -319,6 +320,7 @@ __all__ += [
'bu_latest',
'bu_1_0',
'bu_2_0',
+ 'bu_2_0_mini_preview',
]
# NOTE: OCI backend is optional. The try/except ImportError and conditional __all__ are required
diff --git a/browser_use/tokens/custom_pricing.py b/browser_use/tokens/custom_pricing.py
index 3bb00c905..b625ea758 100644
--- a/browser_use/tokens/custom_pricing.py
+++ b/browser_use/tokens/custom_pricing.py
@@ -27,6 +27,16 @@ CUSTOM_MODEL_PRICING: dict[str, dict[str, Any]] = {
'max_input_tokens': None, # Not specified
'max_output_tokens': None, # Not specified
},
+ 'bu-2-0-mini-preview': {
+ 'input_cost_per_token': 0.15 / 1_000_000, # $0.15 per 1M tokens
+ 'output_cost_per_token': 1.50 / 1_000_000, # $1.50 per 1M tokens
+ # No cache discount on this model: cached reads bill at the input rate.
+ 'cache_read_input_token_cost': 0.15 / 1_000_000, # $0.15 per 1M tokens
+ 'cache_creation_input_token_cost': None, # Not specified
+ 'max_tokens': None, # Not specified
+ 'max_input_tokens': None, # Not specified
+ 'max_output_tokens': None, # Not specified
+ },
'claude-sonnet-4-6': {
'input_cost_per_token': 3.00 / 1_000_000,
'output_cost_per_token': 15.00 / 1_000_000,
diff --git a/examples/beta_agent/basic.py b/examples/beta_agent/basic.py
index 1eed65ca4..bb7f7a8e0 100644
--- a/examples/beta_agent/basic.py
+++ b/examples/beta_agent/basic.py
@@ -23,7 +23,7 @@ async def main() -> None:
agent = Agent(
task=task,
llm=ChatBrowserUse(model='openai/gpt-5.5'),
- # llm=ChatBrowserUse(), # Browser Use's own optimized model (bu-2-0)
+ # llm=ChatBrowserUse(), # Browser Use's own optimized model (bu-2-0-mini-preview)
# llm=ChatOpenAI(model='gpt-5.5'),
# llm=ChatGoogle(model='gemini-3.1-pro-preview'),
# llm=ChatAnthropic(model='claude-opus-4-8'), # Sonnet also works well.
diff --git a/examples/browser/cloud_browser.py b/examples/browser/cloud_browser.py
index 72eebb9a9..f7bce8520 100644
--- a/examples/browser/cloud_browser.py
+++ b/examples/browser/cloud_browser.py
@@ -21,7 +21,7 @@ async def basic():
agent = Agent(
task='Go to github.com/browser-use/browser-use and tell me the star count',
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
browser=browser,
)
@@ -39,7 +39,7 @@ async def full_config():
agent = Agent(
task='go and check my ip address and the location',
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
browser=browser,
)
diff --git a/examples/browser/custom_headers.py b/examples/browser/custom_headers.py
index d1f4b5cd5..7f6ff4636 100644
--- a/examples/browser/custom_headers.py
+++ b/examples/browser/custom_headers.py
@@ -90,7 +90,7 @@ async def main():
'Open https://httpbin.org/headers in two different tabs and extract the full JSON response. '
'Look for the custom headers X-Custom-Auth, X-Request-Source, and X-Trace-Id in the output and compare the results.'
),
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
browser=browser,
)
diff --git a/examples/demo_mode_example.py b/examples/demo_mode_example.py
index b6c540ba3..b0bbe546e 100644
--- a/examples/demo_mode_example.py
+++ b/examples/demo_mode_example.py
@@ -6,7 +6,7 @@ from browser_use import Agent, ChatBrowserUse
async def main() -> None:
agent = Agent(
task='Please find the latest commit on browser-use/browser-use repo and tell me the commit message. Please summarize what it is about.',
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
demo_mode=True,
)
await agent.run(max_steps=5)
diff --git a/examples/features/csv_file_generation.py b/examples/features/csv_file_generation.py
index efecc5480..7119a36c2 100644
--- a/examples/features/csv_file_generation.py
+++ b/examples/features/csv_file_generation.py
@@ -33,7 +33,7 @@ async def main():
'Create a CSV file called "top_cities.csv" with columns: rank, city name, country, population. '
'Make sure to include all cities even if some data is missing — leave those cells empty.'
),
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
)
history = await agent.run()
diff --git a/examples/features/judge_trace.py b/examples/features/judge_trace.py
index c20030041..3826422f3 100644
--- a/examples/features/judge_trace.py
+++ b/examples/features/judge_trace.py
@@ -27,7 +27,7 @@ Round your result to the nearest 1000 hours and do not use any comma separators
async def main():
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
agent = Agent(
task=task,
llm=llm,
diff --git a/examples/features/rerun_history.py b/examples/features/rerun_history.py
index af30cd1c1..6f8fbaa52 100644
--- a/examples/features/rerun_history.py
+++ b/examples/features/rerun_history.py
@@ -59,7 +59,7 @@ async def main():
# Example task to demonstrate history saving and rerunning
history_file = Path('agent_history.json')
task = 'Go to https://browser-use.github.io/stress-tests/challenges/reference-number-form.html and fill the form with example data and submit and extract the refernence number.'
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
# Optional: Use custom LLMs for AI features during rerun
# Uncomment to use a custom LLM:
diff --git a/examples/features/save_as_pdf.py b/examples/features/save_as_pdf.py
index fd5f71658..9b86e024b 100644
--- a/examples/features/save_as_pdf.py
+++ b/examples/features/save_as_pdf.py
@@ -34,7 +34,7 @@ async def main():
'Go to https://news.ycombinator.com and save the front page as a PDF named "hackernews". '
'Then go to https://en.wikipedia.org/wiki/Web_browser and save just that article as a PDF in A4 format.'
),
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
)
history = await agent.run()
diff --git a/examples/getting_started/01_basic_search.py b/examples/getting_started/01_basic_search.py
index a916287ab..ba4f26f4c 100644
--- a/examples/getting_started/01_basic_search.py
+++ b/examples/getting_started/01_basic_search.py
@@ -19,7 +19,7 @@ from browser_use import Agent, ChatBrowserUse
async def main():
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
task = "Search Google for 'what is browser automation' and tell me the top 3 results"
agent = Agent(task=task, llm=llm)
await agent.run()
diff --git a/examples/getting_started/02_form_filling.py b/examples/getting_started/02_form_filling.py
index 97e44d78d..22b59cfb4 100644
--- a/examples/getting_started/02_form_filling.py
+++ b/examples/getting_started/02_form_filling.py
@@ -30,7 +30,7 @@ from browser_use import Agent, ChatBrowserUse
async def main():
# Initialize the model
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
# Define a form filling task
task = """
diff --git a/examples/getting_started/03_data_extraction.py b/examples/getting_started/03_data_extraction.py
index 18c8b51cd..c204967f3 100644
--- a/examples/getting_started/03_data_extraction.py
+++ b/examples/getting_started/03_data_extraction.py
@@ -30,7 +30,7 @@ from browser_use import Agent, ChatBrowserUse
async def main():
# Initialize the model
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
# Define a data extraction task
task = """
diff --git a/examples/getting_started/04_multi_step_task.py b/examples/getting_started/04_multi_step_task.py
index 5855f9087..2bba002d9 100644
--- a/examples/getting_started/04_multi_step_task.py
+++ b/examples/getting_started/04_multi_step_task.py
@@ -30,7 +30,7 @@ from browser_use import Agent, ChatBrowserUse
async def main():
# Initialize the model
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
# Define a multi-step task
task = """
diff --git a/examples/integrations/agentmail/2fa.py b/examples/integrations/agentmail/2fa.py
index 60c1ff83f..3130ad25a 100644
--- a/examples/integrations/agentmail/2fa.py
+++ b/examples/integrations/agentmail/2fa.py
@@ -28,7 +28,7 @@ async def main():
tools = EmailTools(email_client=email_client, inbox=inbox)
# Initialize the LLM for browser-use agent
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
# Set your local browser path
browser = Browser(executable_path='/Applications/Google Chrome.app/Contents/MacOS/Google Chrome')
diff --git a/examples/models/browser_use_llm.py b/examples/models/browser_use_llm.py
index 4e38698e9..3cb441039 100644
--- a/examples/models/browser_use_llm.py
+++ b/examples/models/browser_use_llm.py
@@ -20,12 +20,14 @@ if not os.getenv('BROWSER_USE_API_KEY'):
async def main():
- # `bu-2-0` is the optimized default. ChatBrowserUse can also route to
- # provider-prefixed models (e.g. 'anthropic/claude-sonnet-4-6', 'openai/gpt-5.5',
- # 'google/gemini-3-pro') through the same gateway - see browser_use_provider_models.py.
+ # `bu-2-0-mini-preview` is the optimized default - what a bare `ChatBrowserUse()` gives you.
+ # For the larger premium model pass `model='bu-2-0'` (or 'bu-latest', which tracks it).
+ # ChatBrowserUse can also route to provider-prefixed models (e.g. 'anthropic/claude-sonnet-4-6',
+ # 'openai/gpt-5.5', 'google/gemini-3-pro') through the same gateway - see
+ # browser_use_provider_models.py.
agent = Agent(
task='Find the number of stars of the browser-use repo',
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
)
# Run the agent
diff --git a/examples/sandbox/example.py b/examples/sandbox/example.py
index 565a70035..7182bd2bc 100644
--- a/examples/sandbox/example.py
+++ b/examples/sandbox/example.py
@@ -37,7 +37,7 @@ async def pydantic_example(browser: Browser):
agent = Agent(
"""go and check my ip address and the location. return the result in json format""",
browser=browser,
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
)
res = await agent.run()
diff --git a/examples/sandbox/structured_output.py b/examples/sandbox/structured_output.py
index 14a8c70f5..a4e2fae78 100644
--- a/examples/sandbox/structured_output.py
+++ b/examples/sandbox/structured_output.py
@@ -29,7 +29,7 @@ async def get_ip_location(browser: Browser) -> AgentHistoryList:
agent = Agent(
task='Go to ipinfo.io and extract my IP address and location details (country, city, region)',
browser=browser,
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
output_model_schema=IPLocation,
)
return await agent.run(max_steps=10)
diff --git a/examples/simple.py b/examples/simple.py
index 449237819..ec8b9176c 100644
--- a/examples/simple.py
+++ b/examples/simple.py
@@ -12,6 +12,6 @@ load_dotenv()
agent = Agent(
task='Find the number of stars of the following repos: browser-use, playwright, stagehand, react, nextjs',
- llm=ChatBrowserUse(model='bu-2-0'),
+ llm=ChatBrowserUse(model='bu-2-0-mini-preview'),
)
agent.run_sync()
diff --git a/examples/use-cases/buy_groceries.py b/examples/use-cases/buy_groceries.py
index d2cb6852a..39e4ed8cc 100644
--- a/examples/use-cases/buy_groceries.py
+++ b/examples/use-cases/buy_groceries.py
@@ -24,7 +24,7 @@ class GroceryCart(BaseModel):
async def add_to_cart(items: list[str] = ['milk', 'eggs', 'bread']):
browser = Browser(cdp_url='http://localhost:9222')
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
# Task prompt
task = f"""
diff --git a/examples/use-cases/pcpartpicker.py b/examples/use-cases/pcpartpicker.py
index c332c150f..d6752f04d 100644
--- a/examples/use-cases/pcpartpicker.py
+++ b/examples/use-cases/pcpartpicker.py
@@ -6,7 +6,7 @@ from browser_use import Agent, Browser, ChatBrowserUse, Tools
async def main():
browser = Browser(cdp_url='http://localhost:9222')
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
tools = Tools()
diff --git a/examples/use-cases/phone_comparison.py b/examples/use-cases/phone_comparison.py
index 65157d1c2..a52fa6d25 100644
--- a/examples/use-cases/phone_comparison.py
+++ b/examples/use-cases/phone_comparison.py
@@ -34,7 +34,7 @@ async def find(item: str = 'Used iPhone 12'):
"""
browser = Browser(cdp_url='http://localhost:9222')
- llm = ChatBrowserUse(model='bu-2-0')
+ llm = ChatBrowserUse(model='bu-2-0-mini-preview')
# Task prompt
task = f"""
diff --git a/skills/open-source/references/models.md b/skills/open-source/references/models.md
index ad277d76b..8193b731d 100644
--- a/skills/open-source/references/models.md
+++ b/skills/open-source/references/models.md
@@ -59,8 +59,9 @@ Optimized for browser automation — highest accuracy, fastest speed, lowest tok
```python
from browser_use import Agent, ChatBrowserUse
-llm = ChatBrowserUse() # bu-latest (default)
-llm = ChatBrowserUse(model='bu-2-0') # Premium model
+llm = ChatBrowserUse() # bu-2-0-mini-preview (default)
+llm = ChatBrowserUse(model='bu-2-0-mini-preview') # Default, named explicitly
+llm = ChatBrowserUse(model='bu-2-0') # Premium model ('bu-latest' tracks this)
```
**Env:** `BROWSER_USE_API_KEY` — get at https://cloud.browser-use.com/new-api-key
@@ -68,7 +69,8 @@ llm = ChatBrowserUse(model='bu-2-0') # Premium model
**Models & Pricing (per 1M tokens):**
| Model | Input | Cached | Output |
|-------|-------|--------|--------|
-| bu-1-0 / bu-latest (default) | $0.20 | $0.02 | $2.00 |
+| bu-2-0-mini-preview (default) | $0.15 | $0.15 | $1.50 |
+| bu-1-0 | $0.20 | $0.02 | $2.00 |
| bu-2-0 (premium) | $0.60 | $0.06 | $3.50 |
| browser-use/bu-30b-a3b-preview (OSS) | — | — | — |
diff --git a/tests/ci/models/test_llm_browseruse.py b/tests/ci/models/test_llm_browseruse.py
index 6e602fbeb..bf1aa8c31 100644
--- a/tests/ci/models/test_llm_browseruse.py
+++ b/tests/ci/models/test_llm_browseruse.py
@@ -25,13 +25,13 @@ async def test_browseruse_bu_latest(httpserver):
# --- Model validation -------------------------------------------------------
-def test_default_model_is_bu_2_0():
+def test_default_model_is_bu_2_0_mini_preview():
chat = ChatBrowserUse(api_key=TEST_API_KEY)
- assert chat.model == 'bu-2-0'
+ assert chat.model == 'bu-2-0-mini-preview'
assert chat.provider == 'browser-use'
[email protected]('alias', ['bu-1-0', 'bu-2-0', 'bu-qa-1'])
[email protected]('alias', ['bu-1-0', 'bu-2-0', 'bu-2-0-mini-preview', 'bu-qa-1'])
def test_bu_aliases_are_accepted(alias):
chat = ChatBrowserUse(model=alias, api_key=TEST_API_KEY)
assert chat.model == alias
@@ -39,10 +39,31 @@ def test_bu_aliases_are_accepted(alias):
assert chat.provider == 'browser-use'
-def test_bu_latest_normalizes_to_bu_2_0():
+def test_bu_latest_still_normalizes_to_bu_2_0():
+ """'latest' tracks the stable premium line, so it does NOT follow the preview default."""
chat = ChatBrowserUse(model='bu-latest', api_key=TEST_API_KEY)
assert chat.model == 'bu-2-0'
assert chat.name == 'bu-2-0'
+ assert ChatBrowserUse(api_key=TEST_API_KEY).model != chat.model
+
+
+def test_bu_2_0_mini_preview_is_priced():
+ """The default model must have a pricing entry, or cost tracking silently reports $0."""
+ from browser_use.tokens.custom_pricing import CUSTOM_MODEL_PRICING
+
+ pricing = CUSTOM_MODEL_PRICING['bu-2-0-mini-preview']
+ assert pricing['input_cost_per_token'] > 0
+ assert pricing['output_cost_per_token'] > 0
+ # bu-latest resolves to bu-2-0, not to the preview, so it keeps the premium pricing.
+ assert CUSTOM_MODEL_PRICING['bu-latest'] == CUSTOM_MODEL_PRICING['bu-2-0']
+
+
+def test_llm_models_shortcut_resolves_mini_preview(monkeypatch):
+ """`llm.bu_2_0_mini_preview` must map the underscored name back to the dashed model id."""
+ monkeypatch.setenv('BROWSER_USE_API_KEY', TEST_API_KEY)
+ from browser_use.llm.models import get_llm_by_name
+
+ assert get_llm_by_name('bu_2_0_mini_preview').name == 'bu-2-0-mini-preview'
@pytest.mark.parametrize(