Nox 1.3 Nova
The balance of depth and speed
1.3 Nova sits directly below Astra in the Nox 1.3 family. It is a hybrid-reasoning model: it decides per question whether to run a step-by-step reasoning pass or answer directly, which keeps replies conversational in speed while staying structured in substance.
Nova is the middle of the 1.3 generation. Its defining trait is hybrid reasoning — a per-question routing decision between deliberate step-by-step thinking and a direct reply. The router grants the reasoning pass where its decision calls for decomposition and answers directly otherwise, so no reply carries the latency of an unnecessary reasoning pass.
It reads a context window of about 260,000 tokens and sees images natively, so photo input is handled in the model itself rather than through a separate pipeline. Attached images, long conversations, and large pasted documents stay in context together.
Its headline benchmark figure is public: the foundation engine behind Nova has a reported 90.0% on GPQA-Diamond, the graduate-level science benchmark spanning biology, chemistry, and physics — within three points of the flagship on that test. As with Astra, the figure comes from the engine's developer, not from Nox.
Nova carries the family's full toolset: photo input handled automatically, inline chart generation, connected-app actions, and consent-gated live web access with cited sources. It runs behind Leo, Nox's deterministic safety layer, whose measured recall and false-positive rate are published and identical across every model in the family.
1.3 Nova is available on Pro and MAX, where it is the most capable everyday chat model. A conversation can be handed down to 1.3 Vela mid-thread, with its history kept intact.
Capabilities
- Hybrid reasoning — a per-question decision between step-by-step thinking and a direct answer
- ~260,000-token context window with native image understanding
- A developer-reported 90.0% on GPQA-Diamond — within three points of the flagship
- Inline chart generation and connected-app actions
- Consent-gated live web access with every source cited
Model details
- Architecture — Hybrid reasoning — routes each question to a step-by-step pass or a direct answer
- Context window — ~260,000 tokens
- Reported benchmarks — 90.0% on GPQA-Diamond (graduate-level science) — as reported by the engine's developer
- Reasoning — Selective — reasons step by step only when the routing decision calls for it
- Latency — Conversational — direct answers arrive fast; reasoned answers take moderately longer
- Photo understanding — Yes — native image input; attach photos as usual and Nox handles them automatically
- Live web access — Yes — consent-gated; Nox asks before searching, then cites every source it uses
- Safety layer — Runs behind Leo, Nox's deterministic red-flag detector, on every message
- Available on — Pro and MAX
Availability: Available on Pro and MAX.