Nox changelog — product milestones

Nox ships continuously; this changelog groups the notable user-facing milestones by month rather than listing every deploy. Dates reflect when a capability became broadly available in the app.

November 2026 — Orion 2.5 and Orion Flash 2.5 announced, Vela 2.5 launched, Leo retired

• Released Orion 2.5, the new flagship, and Orion Flash 2.5, the fast Orion 2.5, for public access in Plan mode on Pro and MAX. Both have workspace tools built in (interactive pages, 3D models, code and video, checked in a browser sandbox, plus file reading), health research tools from High thinking up, a staged review of health answers at Ultra (deeper on MAX), and any topic in Plan mode. HealthBenchAI lists Orion 2.5 Ultra at 69.5 length-adjusted on HealthBench Professional, a tool-assisted evaluation with retained-score arithmetic verification, not independent regrading. Orion Flash 2.5 has not been benchmarked yet. Orion 2.1, Orion 2.1 Pro and Orion Flash 2.1 stay available.
• Retired Leo, Nox's dedicated medical-safety layer: the Nox models now give careful, safety-aware health answers on their own, so the Leo strictness levels (relaxed, standard, strict) and the separate AI safety check were removed. Nox still shows an emergency alert with local emergency numbers when a message describes a possible emergency (basic keyword-based detection).
• Launched Vela 2.5, the new everyday model on every plan: it replaces Vela 2.1 as the default for everyday chat and reads attached images directly. Long conversations (requests over 100K tokens) are available on MAX and above.
• Renamed the thinking levels to Low, Medium, High, Extra High, and Ultra (previously Light, Balanced, Extra, Max, and Ultra) and removed Hyper Ultra, the sixth level that Orion 2.1 offered in Plan mode on MAX. Requests for it now run at Ultra; published results measured at that level stay up as dated history.

September 2026 — The Nox 2 model generation

• Launched the Nox 2 generation: Orion 2 upgrades 1.4 Orion with faster overall responses, greater accuracy, and a lower price; the new 300-billion-parameter Orion Flash 2 delivers the fast Orion experience with full thinking depth; and Vela 2 improves accuracy at the same everyday speed.
• Kept Nox 1.3 Astra unchanged for deep reasoning. Orion Flash 2 is available in Plan mode on Pro and MAX and in NOX WORK, without Hyper Ultra; Vela 2 is available on every plan.
• Retired Nox Agent 1.2 and its dedicated research mode. Plan mode with live web access is now the way to get cited research.
• Retired Nox Light 1.1 from consumer products, with Vela 2 as its successor. Light 1.1 remains available to developers through the Nox developer API.

July 2026 — The Nox 1.3 family, expanded safety layer & trust hub

• Launched the Nox 1.3 model family: 1.3 Astra (the new flagship, on MAX), 1.3 Nova (balanced depth and speed, on Pro and MAX), and 1.3 Vela (the fast everyday model on every plan, including Free) — with live “thinking” shown while the deeper models reason.
• Began retiring Nox Light 1.1: 1.3 Vela is its successor on every plan; Light remains available under “More models” during the transition.
• Named Nox's medical-safety system "Leo" — the same deterministic red-flag screening and AI backstop, now with a name you'll see across the app and an intro card the first time you use it.
• Expanded the deterministic red-flag detector to over 100 rules across 70+ acute categories — from cardiac, stroke, and severe breathing difficulty to trauma, pregnancy, pediatric, environmental, toxicology, and mental-health crises — re-measured on a maintained labeled test set (numbers published on the Accuracy page).
• Added an AI safety backstop: when the pattern rules find nothing, a lightweight AI classifier re-reads recent messages to catch dangerous descriptions worded in slang, another language, or indirectly, and surfaces the same seek-care guidance. It only ever adds a safety note, and is evaluated on its own separate test set (kept out of the published detector metrics).
• Made Leo's strictness your choice: a free control right in the chat box lets you pick how far screening goes — Relaxed, Standard, or Strict — while emergency screening always runs at every level (no level turns it off), and Agent and voice conversations always use the strictest setting.
• Region-aware emergency guidance: set your region in Settings and red-flag banners show the correct local emergency number, falling back to universal numbers (911 / 999 / 112) when the region is unknown.
• New public pages: how Nox works, ring-data demo (labeled sample data), illustrative scenarios, consolidated FAQ, accuracy & testing report, privacy whitepaper, reliability report, support & refunds, changelog, and roadmap.
• Accuracy page now renders the committed, machine-generated red-flag detector metrics — the same numbers as the Trust & Transparency page.
• Search-engine and AI-crawler visibility: every new page is prerendered with structured data, mirrored as markdown, and listed in sitemap.xml and llms.txt.

June 2026 — Transparency, connected apps & search visibility

• Trust & Transparency page with published red-flag detector recall and false-positive metrics, governance record, and data-source licensing.
• Nox's App: confirm-first integrations with Google Calendar, Google Docs, Google Drive, and Todoist — Nox only creates new items and always asks before acting.
• Site-wide SEO/GEO work: prerendered public pages, sitemap, llms.txt, markdown mirrors for AI crawlers, and IndexNow submission on every publish.

May 2026 — Voice calls, memory & incognito

• Live voice calls with Nox (real-time speech in many languages), with the same safety layer applied to voice conversations.
• Memory: Nox remembers durable facts you share (allergies, conditions) with full review-and-delete control in Settings.
• Incognito mode: conversations that are never saved.

April 2026 — Deep research & usage allowances

• Nox Agent: deep research with live web sources and cited, openable resource cards.
• Usage allowances across all plans — automatic refresh on every plan, and extra-usage purchases that never expire.
• Enterprise tier for teams — custom usage, shared workspaces, and custom modes.

March 2026 — The safety foundation

• Deterministic red-flag safety layer that screens every message before the AI answers, with an evaluated, committed metrics report.
• Health library and medicine lookup across 100,000+ FDA-listed medicines.
• Photo & label analysis, inline charts, image generation, Analyze My Day, and Daily Check-in.