Nox Orion Flash 2.1 (New)

Fast Orion model · Released in Nox: October 22, 2026 · Page last updated: October 24, 2026

The fast Orion on a new engine — 552B mixture-of-experts, native photo input

Orion Flash 2.1 is the fast version of Orion and replaces Orion Flash 2 on a new engine. It is a 552-billion-parameter mixture-of-experts model with an asymmetric design — 8 billion parameters active per token during prefill and 16 billion during decode — that reads photos natively, runs a deliberate reasoning pass before answering, and keeps a roughly one-million-token context window, while answering faster and consuming less included usage than Orion 2.1.

Orion Flash 2.1 is the fast version of Orion on a new engine: a 552-billion-parameter mixture-of-experts model, 8 billion active per token during prefill and 16 billion during decode, with a roughly one-million-token context window. It keeps Orion's full thinking depth while responding faster and consuming less included usage than Orion 2.1.

It is natively multimodal: photos attached to a message are read by the model itself. It also runs a deliberate internal reasoning pass before answering and supports the same standard thinking controls and answer limits as Orion 2.1, except Hyper Ultra, which remains exclusive to Orion 2.1.

Compared with Orion Flash 2: the engine grows from a 300-billion-parameter model to a roughly 550-billion-parameter (552B) mixture-of-experts model; it reads photos natively; and its context window is roughly one million tokens. Only about 8 billion parameters are active per token during prefill and 16 billion during decode, which is how it stays fast. Accuracy on HealthBench has not yet been re-measured on this engine, so no improvement in score is claimed.

HealthBench Professional: Orion Flash 2, the previous engine under this name, scored 63.4% raw (95% CI 58.7–67.6) and 57.3% length-adjusted on a Nox self-run of all 525 examples on September 27, 2026 (grader as specified by the benchmark, reasoning effort low). That figure has not been re-measured on the Flash 2.1 engine and is shown for continuity only; it is not independently verified.

It carries the Plan mode toolset, including live web access with cited sources. Like every Nox model, it runs behind Leo, Nox's deterministic safety layer.

Orion Flash 2.1 is available only in Plan mode on Pro and MAX, and as a selectable model in NOX WORK. It is not available in everyday chat, to guests, or in voice.

Capabilities

• 552-billion-parameter mixture-of-experts engine — 8 billion active per token in prefill, 16 billion in decode
• Native photo input
• roughly one-million-token context window
• Deliberate internal reasoning pass before every answer
• Standard thinking controls and answer limits shared with Orion 2.1
• Live web access in Plan mode with cited sources

Model details

• Architecture — Mixture-of-experts — 552 billion parameters total; 8 billion active per token during prefill, 16 billion during decode
• Context window — ~1,000,000 tokens
• Reasoning — Deliberate internal reasoning pass before every answer
• Photo understanding — Yes — native; photos are read by the model itself
• Position — The fast version of Orion — Orion's thinking depth at higher speed and lower cost
• HealthBench Professional — Previous engine (Orion Flash 2): 63.4% raw / 57.3% length-adjusted, Nox self-run, all 525 examples, September 27, 2026 — not re-measured on the Flash 2.1 engine; not independently verified
• Thinking levels — Light, Balanced, Extra, Max, and Ultra — Hyper Ultra is exclusive to Orion 2.1
• Live web access — Yes — in Plan mode, with cited sources
• Safety layer — Runs behind Leo, Nox's deterministic red-flag detector, on every message
• Available on — Plan mode on Pro and MAX; selectable in NOX WORK

Availability: Available in Plan mode on Pro and MAX.

Nox Orion Flash 2.1 is the named replacement for Nox Orion Flash 2 (retired on October 22, 2026).

Nox Orion Flash 2.1 — frequently asked questions

What is Nox Orion Flash 2.1?

Nox Orion Flash 2.1 is Nox's fast orion model. Orion Flash 2.1 is the fast version of Orion and replaces Orion Flash 2 on a new engine. It is a 552-billion-parameter mixture-of-experts model with an asymmetric design — 8 billion parameters active per token during prefill and 16 billion during decode — that reads photos natively, runs a deliberate reasoning pass before answering, and keeps a roughly one-million-token context window, while answering faster and consuming less included usage than Orion 2.1.

What can Nox Orion Flash 2.1 do?

Nox Orion Flash 2.1's capabilities: 552-billion-parameter mixture-of-experts engine — 8 billion active per token in prefill, 16 billion in decode; Native photo input; roughly one-million-token context window; Deliberate internal reasoning pass before every answer; Standard thinking controls and answer limits shared with Orion 2.1; Live web access in Plan mode with cited sources.

What architecture does Nox Orion Flash 2.1 use?

Architecture: Mixture-of-experts — 552 billion parameters total; 8 billion active per token during prefill, 16 billion during decode

How large is Nox Orion Flash 2.1's context window?

Context window: ~1,000,000 tokens

Which thinking levels does Nox Orion Flash 2.1 offer?

Thinking levels: Light, Balanced, Extra, Max, and Ultra — Hyper Ultra is exclusive to Orion 2.1

Which Nox plans include Nox Orion Flash 2.1?

Available in Plan mode on Pro and MAX.

Which retired Nox models does Nox Orion Flash 2.1 replace?

Nox Orion Flash 2.1 is the named replacement for Nox Orion Flash 2 (retired on October 22, 2026). Each retired model's page states what its replacement covers.

Related Nox models

• Nox Orion Flash 2 — Retired — the first fast Orion, a 300B model (retired)
• Nox Orion 2.1 — The flagship, now with a peer-review answer pipeline
• Nox Vela 2.1 — The fast model on every plan
• 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