Anthropic's Claude Fable 5 Now Powers Browser Use Terminal's Autonomous Web Agent
Browser Use Terminal v0.1.6 adds Claude Fable 5, Anthropic's new Mythos-class model built for long-horizon agentic tasks with a 1M-token context window
- Browser Use Terminal v0.1.6 now supports Claude Fable 5, Anthropic's new Mythos-class model, available now via a one-line install.
- Claude Fable 5 is Anthropic's most capable public model: 1M-token context, 128k output, 95% on SWE-bench Verified.
- Pricing: $10/M input and $50/M output tokens (2x Opus 4.8); paid Claude subscribers get free access during the introductory window.
- Fable vs. Mythos: same underlying model, but Fable 5 adds safety classifiers and is publicly available; Mythos 5 is restricted to Project Glasswing partners.
- API change: refusals return HTTP 200 with
stop_reason: "refusal", not an error; adaptive thinking is always on and cannot be disabled. - Use cases: long-horizon browser automation, multi-step web workflows, autonomous login/data extraction tasks that previously needed constant human check-ins.
Browser automation just got a serious upgrade. Browser Use Terminal v0.1.6 now ships with support for Claude Fable 5, Anthropic's most capable publicly available model and the first of a new capability class the company calls "Mythos." The demo that accompanied the announcement was characteristically unhinged: the agent spent $7.21 autonomously hunting for a coupon code for gum.
A new tier above Opus
Claude Fable 5 is Anthropic's most capable widely released model, built for the most demanding reasoning and long-horizon agentic work. The name change from the familiar Opus/Sonnet/Haiku naming is intentional. Anthropic moved away from the Opus numbering because this model is capable enough to present significant cyber and biological risk without new safeguards. The named class signals a capability tier, not an increment.
The relationship between Fable and Mythos is worth understanding clearly. Mythos 5 is the raw, more capable configuration of the model. Fable 5 is that same model wrapped in production safeguards so it can be offered for general use. Mythos 5 is restricted to Project Glasswing rather than handed out as a normal API model. Think of Fable 5 as the model you can call today, and Mythos 5 as the same engine with the guard rails removed in narrow, controlled settings.
What the numbers look like
The numbers that matter most are the coding and agentic software-engineering evals: 80.3% on SWE-bench Pro, 95.0% on SWE-bench Verified, 29.3% on FrontierCode Diamond, 72.9% on CursorBench 3.1 at max effort, and 88.0% on Terminal-Bench 2.1. Those are strong numbers, particularly on SWE-bench Verified, which tests whether a model can resolve real GitHub issues end-to-end.
The pattern is consistent: the largest gains appear on long, hard, agentic tasks rather than routine chat. That makes the Browser Use Terminal pairing a natural fit. The terminal is exactly the kind of environment where Fable 5's long-horizon reasoning pays off.
The browser agent angle
Browser Use Terminal is best understood as the browser equivalent of Claude Code or Codex. Coding assistants like Claude Code and Codex work well because they give the model autonomy and a real environment to operate in. Browser Use Terminal brings that same pattern to browser automation, giving an AI agent its own browser environment.
The agent can do quite a lot inside that environment:
- Control Chrome and take screenshots
- Inspect and interact with page elements
- Run JavaScript directly on a page
- Manage login secrets and 2FA without exposing credentials to the model
- Maintain resumable sessions with full history
- Sync cookies from your local Chrome profile
- Use a disposable email inbox for sign-up flows
It is suited for long-running, complex, and asynchronous tasks that previously required frequent human check-ins. It is particularly strong at end-to-end work that would otherwise take a person hours, days, or weeks. It executes well-scoped tasks with few mistakes, automatically self-correcting through verification loops.
Four ways to use it
The terminal ships with four distinct surfaces, so you can pick the one that fits your workflow:
- TUI (interactive terminal app): Launch with
browser, type a task, and watch the agent work in real time. You can interrupt and steer mid-run. - CLI (one-shot commands): Scriptable commands like
browser-use-terminal run-anthropic --model claude-fable-5 "your task"for automation and CI pipelines. - Python API: Use
browser_use.beta.Agentbacked by the same Rust runtime for programmatic control. - Coding assistant plugin: Plug into Claude Code, Codex, or OpenCode via a skill that gives the assistant full browser access.
Getting started is a single curl command:
curl -fsSL https://browser-use.com/terminal/install.sh | sh
browserFrom there, use /model in the TUI to switch to Fable 5, or pass it directly on the CLI:
browser-use-terminal run-anthropic --model claude-fable-5 \
"Log in to the vendor portal and download the latest invoice."The cost reality
Claude Fable 5 and Claude Mythos 5 share the same specs and pricing: a 1M token context window by default, up to 128k output tokens per request, priced at $10 per million input tokens and $50 per million output tokens. That is exactly double Opus 4.8's rates.
The coupon-hunting demo that burned $7.21 is a useful calibration point. Browser tasks can run many steps, each consuming tokens for screenshots, page state, and reasoning. Fable 5 earns the premium when the task is hard enough that a cheaper model takes more attempts or more tokens to finish. On simple tasks, the premium is pure waste. The practical advice: route short, well-defined tasks to Sonnet 4.6, and reach for Fable 5 when the task is genuinely multi-step and ambiguous.
Paid Claude subscribers get free access during the introductory window. On the API side, when Anthropic launched Fable 5 and Mythos 5, rates dropped to $10/$50 for both, a 60% reduction from the Mythos Preview rate.
What changes for API integrations
If you are building on top of Fable 5, there is one behavioral change that will catch you off guard. Claude Fable 5 includes safety classifiers that can decline requests. Claude Mythos 5 does not include these classifiers. When a refusal happens, the API returns a successful HTTP 200 response with stop_reason: "refusal", not an error. You need to handle that case explicitly.
Anthropic built in a fallback mechanism: pass the fallbacks parameter and the API will automatically retry on another Claude model. You also will not be billed for a refused request before any output is generated. For browser automation specifically, Anthropic blocks sensitive categories like cyber-exploitation, bio/chem, and distillation at the model level and reroutes them to the safer Opus 4.8.
There is one more API quirk worth knowing. Adaptive thinking is the only thinking mode on Claude Fable 5 and Claude Mythos 5. It applies whenever the thinking parameter is unset. You cannot disable it. The raw chain of thought is also never returned; you get either a summarized version or nothing, depending on your thinking.display setting.
The bigger picture
The better way to understand Fable 5 is not "Claude got a better benchmark score." It is this: Claude Fable 5 is Anthropic's public test of what happens when frontier models become long-horizon agents instead of chatbots. Pairing it with Browser Use Terminal is the clearest expression of that thesis: a model that can reason across thousands of steps, in a real browser, without hand-holding.
The practical question is whether the agent can stay on task across the kind of messy, multi-page flows that real work involves. The coupon demo suggests it can, even if it occasionally overshoots on cost. As Browser Use Terminal matures and the Fable 5 API stabilizes, the combination is worth watching closely for anyone building autonomous web workflows.