Nox Orion 2.5 (Flagship)

Frontier AI model with built-in tools · Page last updated: November 10, 2026

The new flagship — workspace tools built in, health research tools, and a staged review at Ultra

Orion 2.5 is the flagship of the Nox 2.5 generation, now publicly available in Plan mode on Pro and MAX. It runs the same engine as Orion 2.1 — an 824-billion-parameter mixture-of-experts model, 46 billion active per token, with a roughly one-million-token context window — and the change is in what it does around the answer: the workspace tools are built in, health questions get dedicated research tools from High thinking up and a staged review at Ultra, and in Plan mode it answers any topic. There is no separate Pro version; the tools are part of the model.

Engine: mixture-of-experts, 824 billion parameters total, 46 billion active per token, with a roughly one-million-token context window — the same engine as Orion 2.1, at the same usage rate.

Five thinking levels: Low, Medium, High, Extra High and Ultra. Low and Medium stay as fast and as light on usage as before. Deeper reasoning, more tool lookups and more checking start at High and grow with each level; at Extra High a health question first gets a short case brief, and at Ultra it goes through the staged review described below.

Workspace tools, built in: ask for an interactive page, chart, calculator, 3D model, Python program or MP4 video and Orion builds it, checks that it runs in an isolated browser sandbox, and opens it beside the chat with a live preview, source view, version history, restore and download; follow-up messages edit the same artifact. From High up, an artifact that fails its checks is repaired before it is shown and a screenshot of it is reviewed. You can also attach files to a message and Nox reads them.

Health research tools from High: clinical reference lookups, official drug labels, PubMed literature search, clinical-trial registry lookups and a cited evidence search, with more lookups allowed at each higher level. Low and Medium keep the standard medical lookups, so short questions stay quick.

Staged review at Ultra: a health question at Ultra is answered in stages — a case brief, an evidence step, a draft, an independent audit of the draft, and the final reply, with at most one repair. On MAX the review goes deeper, with a weighted specialist checklist in the brief, a curated drug-label reference library and a longer reasoning budget for the brief and the draft. Only the final reply is shown; the review is never pasted into the answer.

Any topic in Plan mode: in Plan mode, Orion 2.5 answers questions outside health — code, writing, analysis, study — instead of steering back to health. Health answers keep Nox's medical answer standard, and the emergency check still runs before every answer.

Copy box: when a reply contains text you will paste somewhere else (a message to send, an email, a prompt, an answer for a form), Orion 2.5 puts just that text in its own box with a Copy button. One tap copies exactly what is in the box, and the explanation stays outside it.

Nox-run evaluations without web access: MedQA 91.5% at High thinking, GPQA Diamond 86.9% at High thinking, and 39.4% on a 10-task HealthBench Hard sample at Ultra, not a full-benchmark result. Separately, HealthBenchAI lists Orion 2.5 Ultra at 69.5 length-adjusted (raw 72.833) on all 525 HealthBench Professional tasks, with a 95% task-bootstrap interval of 65.81–73.03. The tool-assisted run used live web evidence and a low-effort grader, included 40 development tasks and recovered three failed tasks. HealthBenchAI reports running the original evaluation and later checking retained-score arithmetic, not independent answer regrading or clinical validation. Source: https://healthbenchai.com/benchmarks/professional

Availability: now publicly available in Plan mode on Pro and MAX, with no trusted-organization restriction. Existing plan usage limits apply. Not an everyday-chat model.

Capabilities

• 824-billion-parameter mixture-of-experts model, 46 billion active per token
• ~1,000,000-token context window
• Five thinking levels, Low to Ultra
• Built-in workspace: interactive pages, charts, calculators, 3D models, Python and MP4 video, checked in a browser sandbox, with versions and download
• Reads files attached to a message
• Health research tools from High: clinical references, drug labels, PubMed, clinical trials and cited evidence search
• Staged review of health answers at Ultra, deeper on MAX
• Any topic in Plan mode
• Photo input — up to ten images per message
• Live web access in Plan mode, every source cited
• Copy box: paste-ready text in its own box with a one-tap Copy button

Model details

• Architecture — Mixture-of-experts — 824 billion parameters total, 46 billion active per token; the same engine as Orion 2.1
• Context window — ~1,000,000 tokens
• Reasoning — Deliberate internal reasoning pass before every answer, with thinking time shown
• Thinking levels — Five — Low, Medium, High, Extra High, Ultra
• Answer pipeline — Single pass from Low to High; a short case brief before health answers at Extra High; a staged review at Ultra (brief, evidence, draft, audit, final), deeper on MAX. Only the final reply is shown.
• Workspace — Built in, on the web app: artifacts built and checked in an isolated browser sandbox, versioned and downloadable; attached files are read
• Health tools — From High: clinical references, official drug labels, PubMed, clinical-trial registry and cited evidence search; Low and Medium keep the standard medical lookups
• Topics — Any topic in Plan mode; health answers keep the medical answer standard
• Copy box — Text meant to be pasted elsewhere arrives in its own box with a one-tap Copy button; new in the Nox 2.5 models
• MedQA — 91.5% at High thinking, run by Nox without web access
• GPQA Diamond — 86.9% at High thinking, run by Nox without web access
• HealthBench Hard — 39.4% at Ultra on a subset of tasks, graded with the official rubric method; not a full-benchmark result
• HealthBench Professional — Orion 2.5 Ultra: 69.5 length-adjusted, raw 72.833; 525 tasks, including 40 development tasks; 95% task-bootstrap interval 65.81–73.03. HealthBenchAI's tool-assisted product-pipeline evaluation with live web evidence and three failure recoveries; later arithmetic checks did not independently regrade answers. Not clinical validation. Source: https://healthbenchai.com/benchmarks/professional
• Photo understanding — Yes — up to ten images per message, read automatically
• Live web access — Yes — in Plan mode, every source cited
• Available on — Now publicly available in Plan mode on Pro and MAX; existing usage limits apply

Availability: Now publicly available in Plan mode on Pro and MAX.

Nox Orion 2.5 — frequently asked questions

What is Nox Orion 2.5?

Nox Orion 2.5 is Nox's frontier ai model with built-in tools. Orion 2.5 is the flagship of the Nox 2.5 generation, now publicly available in Plan mode on Pro and MAX. It runs the same engine as Orion 2.1 — an 824-billion-parameter mixture-of-experts model, 46 billion active per token, with a roughly one-million-token context window — and the change is in what it does around the answer: the workspace tools are built in, health questions get dedicated research tools from High thinking up and a staged review at Ultra, and in Plan mode it answers any topic. There is no separate Pro version; the tools are part of the model.

What can Nox Orion 2.5 do?

Nox Orion 2.5's capabilities: 824-billion-parameter mixture-of-experts model, 46 billion active per token; ~1,000,000-token context window; Five thinking levels, Low to Ultra; Built-in workspace: interactive pages, charts, calculators, 3D models, Python and MP4 video, checked in a browser sandbox, with versions and download; Reads files attached to a message; Health research tools from High: clinical references, drug labels, PubMed, clinical trials and cited evidence search; Staged review of health answers at Ultra, deeper on MAX; Any topic in Plan mode; Photo input — up to ten images per message; Live web access in Plan mode, every source cited; Copy box: paste-ready text in its own box with a one-tap Copy button.

What architecture does Nox Orion 2.5 use?

Architecture: Mixture-of-experts — 824 billion parameters total, 46 billion active per token; the same engine as Orion 2.1

How large is Nox Orion 2.5's context window?

Context window: ~1,000,000 tokens

Which thinking levels does Nox Orion 2.5 offer?

Thinking levels: Five — Low, Medium, High, Extra High, Ultra

Which Nox plans include Nox Orion 2.5?

Now publicly available in Plan mode on Pro and MAX.

Related Nox models

• Nox GHI 1 — The officially released model system that powers Pebs, Nox's agent
• Nox Orion Flash 2.5 — The fast Orion 2.5 — native photo input, workspace tools built in, and the same review at Ultra
• Nox Orion 2.6 Cyber Preview — A dedicated security model for defensive and authorized security work — hardened against misuse and prompt injection
• Nox Orion 2.1 — The Nox 2.1 flagship, with a peer-review answer pipeline
• Nox Orion 2.1 Pro — Visualize your body: 3D models, code, and MP4 video
• Nox Orion Flash 2.1 — The fast Orion on a new engine — 552B mixture-of-experts, native photo input
• Nox Vela 2.5 — The fast everyday model on every plan — now reads images
• Nox Voice 2 — The most powerful Nox voice yet — expressive, interruptible, Plan mode in the call

Back to the Nox model family · Plans and pricing · Nox features