Yutori Navigator n2: 27B Computer-Use Model Explained
Yutori Navigator n2 dropped this week, and if you’ve been waiting for a computer-use model that doesn’t cost a fortune per task, the short answer is just below — I’ve dug through the full launch post so you don’t have to.
Short answer:
Navigator n2 is Yutori’s new 27B-parameter computer-use model, announced 26 August 2026.
Unlike n1 and n1.5 (browser-only), n2 operates full desktop environments across Linux, macOS and Windows.
Yutori reports 65.2% on OSWorld 2.0 at $1.46 per task, and 85.3% on OSWorld-Verified — frontier-level scores at a fraction of frontier cost (their numbers, caveat below).
Its party trick: it switches between GUIs, CLIs, tools and code mid-task, using whichever interface finishes the job fastest.
API pricing: $0.5 per million input tokens ($0.05 cached) and $4 per million output tokens.
Yutori Navigator n2: what just dropped
On 26 August 2026, Yutori published “Navigator n2: frontier computer use at a fraction of the cost” on their blog. The pitch: a 27B-parameter model that advances the cost-accuracy Pareto frontier for computer use — in plain English, near-frontier task success at a price that doesn’t make per-task automation silly.
The lineage matters here. Yutori’s earlier Navigator n1 and n1.5 models specialised in browser use — driving websites. n2 is the company’s first general-purpose computer-use model: it operates full desktop environments across Linux, macOS and Windows. That’s native applications, terminals and browsers, all in one agent loop.
Why should an SEO or marketer care? Because computer-use agents are the layer that finally connects “AI can think” to “AI can do” — logging into dashboards, pulling reports, updating spreadsheets, publishing content. It’s the same direction of travel as tools like the Hermes desktop browser and the wider agentic operating system trend I’ve covered — except n2 is a purpose-built model rather than a harness around a general one.
Personalised computer use across a user’s apps and accounts
82.6%
MacAgentBench
Real-world macOS tasks across 25 applications
83.1%
WeaveBench
Tasks combining GUI interaction with CLI and code execution
70.3%
In Yutori’s own comparison charts, n2 sits just behind Claude Opus 5 on OSWorld 2.0 (70.6% vs 65.2%) while beating the GPT-5.6 variants they list, and it tops their OSWorld-Verified, MyPCBench, MacAgentBench and WeaveBench tables outright.
⚠️ Caveat: every number above is Yutori’s self-reported benchmarking from their 26 August 2026 launch post, including the competitor comparisons. Vendor-run evals routinely flatter the vendor. Wait for independent OSWorld leaderboard verification before treating the exact rankings as settled.
🔥 Want this set up without the guesswork? Inside the AI Profit Boardroom we’re testing computer-use agents like Navigator n2 on real SEO workflows — 3,700+ members, four live calls per week, daily tutorials, done-for-you templates and a 30-day roadmap. Or if you’d rather map out where agent automation fits your business first, book a free SEO strategy session and we’ll build the plan together.
What makes Yutori Navigator n2 different: GUIs, CLIs and code
The design insight behind n2 is the bit I find most interesting. Humans use computers through interfaces designed for humans — but computer-use models don’t have to. n2 is trained to interleave every interface a computer offers: it clicks through an application when visual interaction is needed, drops into the shell for system operations, calls tools for structured data, and writes short snippets of code when that’s faster.
Yutori’s own example nails why this matters: renaming 100 photos from IMG_1337.jpg to paris-trip-001.jpg is one for-loop in the shell, versus 100 rounds of right-click → rename → type → enter in a GUI. A GUI-only agent burns time and tokens doing things the slow, human way. Equally, plenty of real-world tasks can’t be reduced to APIs — native apps, live websites and visual states still need actual clicking. n2 does whichever is fastest, per step, inside a single task.
For anyone building automation this is the practical difference between an agent that costs pennies per completed workflow and one that costs pounds. It’s also why I think multimodal agent models — like the vision-enabled experiments I covered in my DeepSeek V4 Flash Vision breakdown — are converging on the same “see the screen, but prefer the terminal” pattern.
Pricing and how to get started
API pricing from the launch post: $0.5 per million input tokens ($0.05 per million cached) and $4 per million output tokens. That’s what produces the $1.46-per-task OSWorld 2.0 figure Yutori quotes, and it undercuts the frontier models in their comparison chart by an order of magnitude on cost per task.
Access is through Yutori’s API (docs at docs.yutori.com) alongside their consumer products — Delegate, Scouts and Local. There’s no open-weights release announced in the launch post, so unlike the open models I usually test locally, this one lives behind the API for now.
My suggested first experiment: pick one repetitive, dashboard-heavy task in your business — reporting, citation checks, listing updates — and price out what n2 would cost per run versus your current VA or manual time. At $1.46-a-task economics, plenty of “not worth automating” jobs suddenly are.
The bottom line on Yutori Navigator n2
Navigator n2 is the most credible attempt yet at making computer-use agents cheap enough for everyday business automation. The scores are Yutori’s own, so hold them loosely — but the pricing is public, the Linux/macOS/Windows scope is a real step past browser-only agents, and the interleaved GUI/CLI/code approach is clearly the right architecture. If the independent leaderboards confirm even most of the launch numbers, this becomes the default cost-effective pick for agent workflows.
Want help working out which of your workflows are worth handing to an agent first? Book a free SEO strategy session and I’ll walk through it with you.
FAQ: yutori navigator n2
What is Yutori Navigator n2?
Navigator n2 is a 27-billion-parameter computer-use model from Yutori, announced on 26 August 2026. It operates full desktop environments across Linux, macOS and Windows, switching between GUIs, command lines, tools and code within a single task.
How is n2 different from Navigator n1 and n1.5?
n1 and n1.5 specialise in browser use — driving websites. n2 is a general-purpose computer-use model that can drive full desktop environments: native apps, terminals and browsers, not just the web.
How much does Yutori Navigator n2 cost?
Yutori’s API pricing is $0.5 per million input tokens, $0.05 per million cached input tokens, and $4 per million output tokens. On OSWorld 2.0 that works out, per Yutori, at about $1.46 per task.
What are Navigator n2’s benchmark scores?
Per Yutori’s launch post: 65.2% on OSWorld 2.0, 85.3% on OSWorld-Verified, 82.6% on MyPCBench, 83.1% on MacAgentBench and 70.3% on WeaveBench. All self-reported.
Is Yutori Navigator n2 open source?
No open-weights or open-source licence was announced in the launch post. n2 is accessed through Yutori’s API and their products.
Does Navigator n2 work on Windows and macOS?
Yes — Yutori says n2 operates full desktop environments across Linux, macOS and Windows, and MacAgentBench specifically covers real-world macOS tasks spanning 25 applications.
Next steps: computer-use agents are moving fast, and the people winning are the ones testing them on real workflows now. Join 3,700+ members in the AI Profit Boardroom for four live calls a week, daily tutorials and done-for-you templates — or book a free SEO strategy session and I’ll help you pick your first automation target.
About the author: Julian Goldie is an SEO agency owner with 394K+ YouTube subscribers, a 100% Upwork job-success score, 75K+ community members across his groups, 10+ years in SEO and a best-selling SEO book. Watch his agent experiments on YouTube, learn AI SEO with 3,700+ members in the AI Profit Boardroom, or book a free SEO strategy session. For agency work, book a call for a custom quote.
Last updated August 2026. This is the living guide to yutori navigator n2 — it gets updated as the tools change.
