LATEST
AI Tools

Browser Use 1.0.6 Makes BU 4.0 the Default — at 15x the Price

Updated November 16, 2026
13 min read
Editor's Note: This article was human-reviewed for accuracy
Laptop screen showing a code editor with debugging software, glasses resting in front of the keyboard

Quick Answer & Key Takeaways

  • • Browser Use 1.0.6 makes BU 4.0 the default model for its Python agent library, replacing bu-fast.
  • • BU 4.0 costs $1.80 per million input tokens and $9.00 per million output tokens, 15 times the old default's rates.
  • • Anyone who leaves the model unset pays the new price, but passing llm='bu-fast' keeps the old one.

Upgrade Browser Use this week and your agent bill can jump without a single line of code changing. Version 1.0.6, released November 16, 2026, changes the default model for the open-source Python library: Agent(task) and ChatBrowserUse() now run BU 4.0 on the Browser Use gateway instead of bu-fast. The new model costs $1.80 per million input tokens and $9.00 per million output tokens. The old default cost $0.12 and $0.60. That is 15 times as much on both rates, and the release notes say so plainly: "This costs more if you pass no llm." This article works through the new price list, who actually pays more, what Browser Use has not published about BU 4.0, and the one-line change that keeps the old rate.

What changed in Browser Use 1.0.6

Browser Use is an open-source Python library that lets a language model drive a real Chrome browser: it clicks, types, scrolls, and fills in forms on your behalf. The project, MIT-licensed with 121,200 stars on GitHub as of November 16, 2026, also sells a hosted cloud with stealth browsers and an agent API, all billed from one prepaid credit balance. Version 1.0 arrived earlier in November with a ground-up rewrite in which the agent writes Python in a persistent read-eval-print loop connected to Chrome through the Chrome DevTools Protocol. Version 1.0.6, published November 16, is a small release with one large billing consequence.

The release notes list four changes:

  • BU 4.0 is the default model. Agent(task) and ChatBrowserUse() now run bu-4.0 on the Browser Use gateway instead of bu-fast. The notes point users at llm='bu-4.0-mini' for simple tasks at the lowest cost in the new family, or llm='bu-4.0-pro' for the hardest tasks.
  • The default reasoning level rises. The new default runs at high reasoning instead of medium, which is part of why tasks cost more: the model spends more effort on each step.
  • First-run login is smoother. The first run without an API key opens Browser Use login and continues after approval, and cloud browsers print their live-view link.
  • The built-in judge stays grounded. The judge that verifies task completion now sticks to what the run actually observed.

The same switch reaches the hosted service. Browser Use's own documentation says that in the US, projects created on or after November 16, 2026 that omit the model field use BU 4.0, and users of the US web app without a saved model choice start on BU 4.0. So the change is not only a Python default. It is the default for new cloud projects too.

The new price, in numbers

Here is the current model price list from Browser Use's own pricing page, in US dollars per million tokens:

ModelInputCached inputOutput
BU 4.0 Mini (bu-4.0-mini)$0.20$0.02$1.00
BU 4.0 (bu-4.0) — new default$1.80$0.09$9.00
BU 4.0 Pro (bu-4.0-pro)$3.00$0.15$15.00
BU Fast (bu-fast) — old default$0.12$0.012$0.60
BU Ultrafast (bu-ultrafast)$0.24$0.024$1.20
GPT-6 Luna (gpt-6-luna)$0.12$0.012$0.60

Two details in that table deserve attention. First, the 15-times figure applies to fresh input and output tokens: $1.80 against $0.12, and $9.00 against $0.60. Cached input is different. BU 4.0 charges $0.09 per million cached input tokens against $0.012 for bu-fast, which is 7.5 times as much, not 15. Cache-heavy workloads therefore see a smaller jump than the headline suggests. Second, BU 4.0 Mini, the cheapest model in the new family, still costs more than the old default on both rates: $0.20 and $1.00 against $0.12 and $0.60. Switching to Mini softens the increase. It does not remove it.

For readers outside the US, the same rates convert to roughly £1.36 per million input tokens and £6.82 per million output tokens for BU 4.0, or about C$2.57 and C$12.86. The old bu-fast rates work out to about £0.09 and £0.45, or C$0.17 and C$0.86. Those conversions use November 16, 2026 mid-market rates of 1 USD = £0.7576 and C$1.4289, and they are approximations: Browser Use bills in US dollars from a prepaid credit balance, where 1 credit equals $1.

What 15 times actually means for a task

A rate is not a bill. What you pay depends on how many tokens a task uses, and BU 4.0's higher reasoning level means a task may use more tokens per step than it did on bu-fast. Browser Use has published no token-count comparison between the two models, so the honest way to size the change is to run the arithmetic on a fixed workload and treat the vendor's own example as a single data point.

Take a task that uses 100,000 input tokens and 10,000 output tokens. On bu-fast, that costs 0.1 × $0.12 plus 0.01 × $0.60, which is $0.012 + $0.006 = $0.018. On BU 4.0, the same tokens cost 0.1 × $1.80 plus 0.01 × $9.00, which is $0.18 + $0.09 = $0.27. Same tokens, 15 times the price. Browser Use's own example in the release notes tells a similar story: one short test task cost $0.022 on BU 4.0 instead of $0.0014 on bu-fast, a gap of about 15.7 times. Treat that as one vendor example, not a forecast for your workload: your tasks will use different token counts, and the reasoning change can push the gap wider.

Scale that example up and the stakes become clear. Ten thousand tasks a month at the vendor's example cost would run about $14 on bu-fast and about $220 on BU 4.0. That is arithmetic on the release notes' single example, not a measured average, but it shows why the default matters: a dependency upgrade that nobody reviews can move a monthly bill from a rounding error to a line item. The parts of the bill that do not change are browser time and network traffic, which stay flat, so the total increase for a real workload will be less than 15 times overall. How much less depends on how much of your spend was model tokens in the first place.

What BU 4.0 is, and what Browser Use has not said

BU 4.0 is Browser Use's own model family, offered in three sizes. The documentation describes BU 4.0 Mini as fast and low cost, BU 4.0 as the best balance, and BU 4.0 Pro as the most capable. All three share one adjustable reasoning setting, with high as the default, and all three run on standard processing only, with no fast or priority tiers. Model IDs stay pinned, with no silent upgrades: future generations will use versioned IDs such as bu-5.0.

What the project has not said is just as important. Browser Use has not disclosed what BU 4.0 is built on. That stands out because the old default is fully identified: the project's own documentation says bu-fast is GPT-6 Luna at medium reasoning, and bu-ultrafast is the same model at low reasoning on the priority tier. For BU 4.0 there is no equivalent statement, so do not assume it is a fine-tuned version of any particular model.

Browser Use has also not published a benchmark score for BU 4.0. Its public benchmark board, last updated October 29, 2026 with 180 tasks, ranks GPT-6.1 Sol first at 81.3, ahead of GPT-6 Sol at 79.4 and GPT-6.1 Terra at 79.0. BU 4.0 does not appear on that board, which predates the new model. The claim that BU 4.0 completes tasks better is the project's own description, not a published number. If BU 4.0 is worth 15 times the token price for your work, that is something to verify on your own tasks before moving production workloads to it, not something the release notes prove.

Two more limits are worth knowing. BU 4.0 and BU 4.0 Pro are US-only for now; only Mini is available in the US and EU. And BU 4.0 models always run on Browser Use's platform route, even when a project has its own provider keys. Bring-your-own-key pricing, where you pay the provider directly plus a 0.2-times Browser Use orchestration fee, does not apply to them.

Who pays more, and who does not

The price rise is narrow in one sense: it only touches people who leave the model unset. But that is exactly what most quick-start code and most new cloud projects do, so the affected group is large.

You pay the new rate if you upgrade to browser-use 1.0.6 or later and keep calling Agent(task) or ChatBrowserUse() with no llm argument, if you create a new US cloud project on or after November 16, 2026 without setting a model, or if you use the US web app with no saved model choice. You do not pay it if you set a model explicitly, if your US project predates the change (older US projects keep GPT-6 Luna with high reasoning when model and model parameters are omitted), if you are in the EU (defaults there are unchanged), or if your project is Bedrock-only, which keeps its compatible default.

The free plan softens the landing a little. BU 4.0 and BU 4.0 Mini are available on the free plan, so a new user can try the default without buying credits first. BU 4.0 Pro needs credits. Browser Use also gives every new signup a one-time $1 credit that never expires, and top-ups start at $5. That $1 covers roughly 45 of the vendor's example tasks on BU 4.0, or about 700 on bu-fast, which is another way of seeing the size of the gap.

Browser Use is far from the only agent you can run yourself. If you want an agent that lives on your own machine rather than in a browser tab, n8n Nod is a free open-source personal agent with its own computer, and OpenBot 1.0 offers free AI coworkers with a $29 team tier. Neither charges per token the way a hosted gateway model does, which is the trade-off this whole release turns on.

The rest of the bill has not changed

Model tokens are one of three meters on a Browser Use bill, and the other two are untouched by 1.0.6. Cloud browsers cost $0.06 per browser hour, billed by the minute with a one-minute minimum. Network traffic costs $5 per gigabyte through Browser Use's home-IP proxies, or $0.20 per gigabyte if you bring your own proxy or go direct. CAPTCHA solving is included. There is no subscription and no concurrency fee: you pay as you go from prepaid credits, and top-ups never expire. Concurrency rises with total paid, from 10 browsers at once to start up to 1,000 once you have paid $25,000 in total.

That structure is why the headline number needs context. If your workload is browser-heavy and token-light, for example long sessions on media-rich pages with little reasoning, the model switch may barely move your bill. If your workload is token-heavy, with long pages and many reasoning steps per task, the model is most of the cost and the increase lands almost in full. Check result.cost_usd after a run: the project's documentation describes it as what the gateway actually charged, which makes it the fastest way to see the change in your own numbers.

How to keep the old price

The fix is one argument. Pass llm='bu-fast' when you create your agent, and you keep the old model and the old rates:

from browser_use import Agent

agent = Agent(task="Summarise the pricing page", llm='bu-fast')
await agent.run()

For simple tasks where the new family is acceptable, llm='bu-4.0-mini' is the cheapest BU 4.0 option at $0.20 per million input tokens and $1.00 per million output tokens, though it still costs more than bu-fast. For the hardest tasks, llm='bu-4.0-pro' costs $3.00 and $15.00. In cloud projects, set the model explicitly instead of leaving it blank, and in the web app save a model choice so the new default does not apply.

Pin the version, too. The project itself advises pinning an exact version, such as browser-use==1.0.6, never a floating range, because releases move fast and can change defaults. That is the same discipline that applies to terminal coding agents: if you build agents that write code, Claude Code is Anthropic's agentic coding tool, explained here. Pin the version, read the release notes, then upgrade on purpose. A dependency upgrade should never be the moment you discover a price change.

Common mistakes to avoid

  • Upgrading with a floating version and no model set. A >= requirement pulls in 1.0.6 silently, and every agent without an llm argument starts billing at BU 4.0 rates. Pin the version and set the model.
  • Reading "15 times" as "my bill will be 15 times bigger." The rate is 15 times. Your bill depends on your token mix: cached input is only 7.5 times as much, and browser hours and traffic do not change at all.
  • Assuming the change applies everywhere. It does not. EU defaults are unchanged, older US projects keep GPT-6 Luna, and Bedrock-only projects keep their own default.
  • Assuming your own API key covers BU 4.0. It does not. BU 4.0 models always use Browser Use's platform route, so a bring-your-own-key setup does not lower their price.
  • Assuming BU 4.0 is proven better. No benchmark score for BU 4.0 has been published. Run your own tasks on both models and compare success rates and result.cost_usd before switching production work.
  • Forgetting the reasoning change. The default moved from medium to high reasoning along with the model. Even at the same rates, that would mean more tokens per step. At 15 times the rates, it is worth watching.

Frequently asked questions

Does Browser Use 1.0.6 make every task 15 times more expensive?

No. The per-token rates for BU 4.0 are 15 times the bu-fast rates, but what a task costs depends on how many tokens it uses. Browser Use's own example task cost $0.022 on BU 4.0 against $0.0014 on bu-fast, about 15.7 times as much. Cached input tokens cost only 7.5 times as much, and browser time and traffic are unchanged, so the increase on a full bill is smaller than 15 times.

How do I keep using bu-fast after upgrading?

Pass llm='bu-fast' when you create the agent, as shown above. That keeps the old model and the old rates of $0.12 per million input tokens and $0.60 per million output tokens. In cloud projects, set the model explicitly instead of leaving the field blank.

Is BU 4.0 available in Europe?

Only BU 4.0 Mini is available in the EU for now. BU 4.0 and BU 4.0 Pro are US-only, and EU project defaults are unchanged, still GPT-6 Luna with high reasoning.

Can I use my own API key with BU 4.0?

No. BU 4.0 models always run on Browser Use's platform route, even when a project has its own provider keys. Bring-your-own-key pricing, paying the provider directly plus a 0.2-times orchestration fee, applies to other models but not to the BU 4.0 family.

Is Browser Use still free?

The Python library is free and MIT-licensed, and you can run it against your own Chrome at no charge beyond your model provider's bill. The hosted side is pay as you go: cloud browsers at $0.06 an hour, traffic from $0.20 per gigabyte, and gateway model tokens at the published rates. New signups get a one-time $1 credit that never expires.

What is the cheapest way to run a Browser Use agent now?

bu-fast is tied with GPT-6 Luna as the cheapest listed per-token model at $0.12 per million input tokens and $0.60 per million output tokens, but only if you set it explicitly. BU Ultrafast is quoted at $0.006 per task on the pricing page. The new default, BU 4.0, costs far more per token than either; only BU 4.0 Pro, at $3.00 per million input tokens and $15.00 per million output tokens, costs more.

Conclusion

Browser Use 1.0.6 is a small release with a large billing consequence: the default model is now BU 4.0, at 15 times the token price of bu-fast, with higher reasoning as the default. Nothing about the library, the license, or the other meters has changed, and the old price is one argument away. Set llm='bu-fast' before your next deploy, check result.cost_usd on the first run after upgrading, and read the release notes before the next version bump. The cheapest lesson here is the one you learn from the notes, not from the invoice.

Sources

  • Browser Use 1.0.6 release notes — GitHub, November 16, 2026: BU 4.0 becomes the default model; 15-times token prices; $0.022 against $0.0014 example task.
  • Browser Use pricing — official pricing page: BU 4.0 Mini $0.20/$1.00, BU 4.0 $1.80/$9.00, BU 4.0 Pro $3.00/$15.00, BU Fast $0.12/$0.60 per million tokens; $0.06 per browser hour; $5/GB home-IP traffic; $1 signup credit.
  • Browser Use cloud models documentation — US projects created on or after November 16, 2026 default to BU 4.0; older US projects keep GPT-6 Luna; EU defaults unchanged; BU 4.0 and Pro US-only; no bring-your-own-key for BU 4.0 models.
  • Browser Use open-source models documentation — bu-fast is GPT-6 Luna at medium reasoning; bu-ultrafast is the same model at low reasoning.
  • Browser Use benchmarks — public benchmark board, October 29, 2026: GPT-6.1 Sol 81.3 first of 180 tasks; no BU 4.0 score published.
  • Browser Use on GitHub — 121,200 stars, 13,551 forks, MIT license, verified via the GitHub API on November 16, 2026.
  • Currencyflow mid-market exchange rates — November 16, 2026: 1 USD = £0.7576 and C$1.4289, used for the GBP and CAD conversions in this article.
13 min read
4 views
edit_note Written by

Ratul Ahamed

Founder & Editor-in-Chief

Ratul Ahamed is the founder and editor of AI Core Updates, an independent site covering AI tools, models and pricing. He is a web developer based in Bangladesh and also runs All In One Tools, a free collection of browser-based utilities. Contact: [email protected]

Read all stories by Ratul Ahamed arrow_forward
verified_user Human-Reviewed Editorial

forum Comments (0)

Leave a Reply

chat_bubble_outline

No comments yet. Be the first to join the conversation!

Weekly AI Digest

Get our weekly AI tools digest

One email a week analyzing new AI tools, model updates, and practical workflows. Tested and summarized with zero fluff. Unsubscribe any time.

bolt
Hands-on AI Tool Reviews
Pricing math, limits, and alternatives.
verified_user
Strict Anti-Spam Policy
1-click unsubscribe at any time.
mark_email_unread

Subscribe to Digest

Sent weekly • No spam

person_outline
mail_outline

No spam ever. You can unsubscribe with a single click at any time.

cookie

Cookie Settings

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. Learn more