
GPT-6.2 Astra Pro Is Not a Model: What Shipped on 17 November, and What "Pro" Actually Turns On
- GoogleNEWGoogle: Gemini 4 Carbon2026-11-10$2.00 / $10.00 per 1M tokens · 90 tok/s
- AnthropicNEWAnthropic: Claude Fable 5.22026-11-09$8.00 / $40.00 per 1M tokens · 97 tok/s
- orcaNEWOrca: OrcaC2 1.02026-11-0711 tok/s
- OrcaNEWOrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 per 1M tokens · 88 tok/s
- OrcaNEWOrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 per 1M tokens · 133 tok/s
- OrcaNEWOrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$4.00 / $20.00 per 1M tokens · 86 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 per 1M tokens · 124 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 per 1M tokens · 67 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 91 tok/s
- OrcaOrcaCyber: MiMo V2.6 Flash Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 43 tok/s
- MoonshotMoonshotAI: Kimi K3.1 Pro2026-10-2447Intelligence
- OrcaOrcaCyber: DeepSeek V4.1 Flash Abliterated / Uncensored2026-10-23$0.80 / $1.60 per 1M tokens · 192 tok/s
- minimaxMiniMax: M3.1 Flash2026-10-1929Intelligence
- OrcaOrca: OrcaCyber Ultra 1.02026-10-17$5.00 / $10.00 per 1M tokens · 133 tok/s
- deepseekDeepSeek: DeepSeek V4.1 Flash 10172026-10-1733Intelligence
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740Intelligence
- OrcaOrca: OrcaCyber Zero 1.52026-10-10$3.00 / $7.50 per 1M tokens · 131 tok/s
- openaiOpenAI: GPT-6.1 Sol2026-09-2955Intelligence
- anthropicAnthropic: Claude Sonnet 5.52026-09-2855Intelligence
- typesafeTypeSafe: Jev 1.132026-09-24$0.04 / $0.00 per 1M tokens · 186 tok/s
Ask OpenAI's documentation for gpt-6.2-astra-pro and you get a 404. There is no model page for that string, no row for it in the published rate card, and no line for it in the 17 November changelog. Yet GPT-6.2 Astra Pro is the name a lot of people are typing into search this week, because that is the label model marketplaces print when they list GPT-6.2 Astra with reasoning.mode set to pro instead of the default standard. The model underneath the label is real and it is one day old: OpenAI released GPT-6.2 Astra, alongside GPT-6.2 Sol, on 17 November 2026, at $8 per million input tokens and $40 per million output. "Pro" is a switch on that model. The switch has a bill attached, and the bill is the part worth understanding before you flip it.
What shipped on 17 November, precisely
The changelog entry is short and unambiguous: two new model IDs, gpt-6.2-astra and gpt-6.2-sol, both available on the Responses API and Chat Completions, with tool calling requiring Responses. The launch post adds the consumer side — Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex from day one, and Ultrafast inference available on the API for both models at launch. For the first time, Ultrafast also reaches Pro 200 subscribers, starting with GPT-6.2 Sol.
What you can call, then, is two strings. Here is the shape of each, from OpenAI's own model pages:
• gpt-6.2-astra — 1,050,000-token context window, 922,000 maximum input, 128,000 maximum output; text and image in, text out; knowledge cutoff 18 May 2026; effort levels low, medium, high, xhigh, max, with none rejected outright.
• gpt-6.2-sol — same 1,050,000 / 922,000 / 128,000 envelope, cutoff 1 June 2026, and the same rejection of none.
• Rate card, short context — $8 input, $0.40 cached input, $10 cache write for 30 minutes of retention, $40 output per million tokens on Astra; $2 / $0.10 / $2.50 / $10 on Sol.
• Rate card, above 272,000 input tokens — input and cache rates double and output goes to 1.5× for the whole request, so Astra bills $16 / $0.80 / $20 / $60.
• Tiers — Batch at 50% of standard, Fast mode at 2×, Ultrafast at 3× through 28 February 2027 and 6× from 1 March 2027.

Nothing in that list is called Pro. The label is not OpenAI's, and that distinction is not pedantry — it is the difference between a model you can look up and a parameter you have to pass.
Why a mode gets sold as a model
OpenAI introduced reasoning.mode with the GPT-5.6 generation, and the docker-style phrasing of the API reference explains the confusion in a single sentence: "GPT-5.6 and GPT-6 models support standard and pro reasoning modes in the Responses API. standard is the default."
Before that, Pro was a separate identifier. GPT-5.2 Pro, GPT-5.4 Pro and GPT-5.5 Pro are distinct strings with their own model pages and their own prices — $30 in and $180 out for GPT-5.4 Pro, against $2.50 and $15 for the base model. When a marketplace needed to expose pro reasoning on a GPT-6 model, it had no separate ID to point at, so it invented one following the old naming convention. That is why gpt-6.2-astra-pro resolves on a catalogue and 404s on the vendor's own docs. The docs are right; the catalogue is describing a configuration, not a checkpoint.
OpenAI keeps both mechanisms alive and says so explicitly: "Existing Pro model IDs keep their current behavior and pricing." So "Pro" now means two different things depending on which string you name. On GPT-5.5 it means a bigger model sold at a premium per token. On GPT-6.2 Astra it means more model work bought at the same per-token rate. Confusing the two is how teams end up budgeting for a price tag that does not exist.
What the switch actually does to a request
Pro mode is not a bigger model and it is not a higher effort level. The docs are careful about this: "Mode selects standard or pro execution, while reasoning.effort controls how much reasoning the model applies within that mode." They are independent axes. You can run pro mode at low effort, and you can run standard mode at max effort.
What pro execution does is aggregate several passes of model work into one answer. The API reference says it "aggregates the model work performed to produce the final answer and bills those tokens at the selected model's standard token rates." No premium rate, no separate line item — just a larger number of tokens crossing the same meter.
That has a concrete consequence for anyone trying to price a workload in advance: you cannot compute the cost of pro mode from the rate card. The multiplier is how much extra work the model decides to do, which varies by task. On a fixed 60,000-token input with a 3,000-token answer, standard mode is $0.48 of input and $0.12 of output — sixty cents. Run the same request in pro mode and the dollar-per-million figures do not move, but the token count does, and nothing in the documentation tells you by how much. The only honest way to budget is to measure your own calls, or to borrow an independent per-task figure.
What the independent measurement says
Artificial Analysis scores GPT-6.2 Astra at 62 on its Intelligence Index (v5.0.1), third of 87 models on the board, up from 57 for GPT-6 Astra. That is a genuinely independent number — the index runs nine evaluations, including Terminal-Bench 4.0 and FrontierCode-AA 2.0, on its own hardware rather than trusting the vendor's harness.
The number that matters for this article is the one next to it: $4.00 per Intelligence Index task. That is the weighted cost of the input, cache, reasoning and answer tokens it took to complete one task, and it is the closest thing to a public estimate of what a hard request costs on this model. For comparison, Artificial Analysis measures $6.24 per task for GPT-6 Astra and $1.27 for GPT-6.1 Sol.
One caution that the board itself invites: Artificial Analysis does not publish a time-per-task figure for GPT-6.2 Astra yet — the panel reads N/A — and its per-task cost is tied to the specific configuration it scored. It does not say whether that configuration used pro mode. Treat $4.00 as the cost of a hard task on the model, not as the price of the switch.

The lever that is cheaper than Pro
Buried in the same 17 November changelog is a discount that applies to GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna, and it is worth more than the mode debate for anyone whose work is not interactive. The Responses API public beta now offers a two-hour or eight-hour background completion window, with token rates 30% lower for the two-hour window and 40% lower for the eight-hour one.
Forty percent off, applied to the whole request, is a bigger saving than most teams will get from fiddling with effort levels — and unlike pro mode, it is exactly computable in advance. If a job can wait, the background window is the cheapest thing on this rate card. If a job cannot wait, pro mode is the opposite trade: you are buying more compute, not less, and paying for it in tokens.
Who should set the switch, and who should not
The case for pro mode is narrow and specific. It earns its cost when the task has a single hard deliverable whose errors are expensive — a migration plan, a contract review, a valuation model — and when latency is not the constraint. That is the same territory where OpenAI's own launch material concentrates: the legal benchmark, the analyst agent, the spreadsheet that has to reconcile three statements. Those are tasks where one better answer is worth several more minutes of compute.
The case against is almost everything else. Classification, extraction, summarisation, routine code edits, anything in a loop — these are volume tasks, and pro mode multiplies the volume. The GPT-6.2 Sol comparison in OpenAI's own launch post makes the point from the other direction: on Cognition's FrontierCode 2.0 at max effort, Sol lands within 0.5 points of Astra at about one-fifth the cost. If a cheaper model is within half a point on your benchmark, spending more per token on the expensive one, and then multiplying that by pro mode, needs a justification that starts with your own evaluation rather than the leaderboard.
How to call it, and one route that does not exist yet
In the Responses API the switch is a nested object on an otherwise ordinary request: model gpt-6.2-astra, reasoning set to {"mode": "pro", "effort": "medium"}, and your input. Chat Completions does not carry the field; tool calling requires Responses anyway, so a pro-mode workload with tools is a Responses workload by construction.
Where you call it from is a separate question. GPT-6.2 Astra is available on the OpenAI API and in ChatGPT Work and Codex, and that is currently the whole list.

It is not on our catalogue yet — OrcaRouter passes provider list price through with 0% markup, which is why the GPT-6 models we do route, GPT-6 Astra at $10 in and $50 out and GPT-6.1 Sol at $2 and $10, are priced at exactly what OpenAI charges rather than at a markup of it. When a model is not in the catalogue, the honest answer is the vendor's own API, and that is where this one is today.
What to watch next
Three things will settle the Pro question better than any rate-card arithmetic. Artificial Analysis publishing a time-per-task number for GPT-6.2 Astra, which would turn the $4.00 into a cost-per-minute as well as a cost-per-task. A pro-mode breakout on that board, so the switch gets its own measured price instead of an inferred one. And OpenAI's promise to bring the 6.2 models to Daybreak — the trusted-access programme that currently carries GPT-6 Astra with reduced cyber refusals — because until that lands, the tightened safeguards on 6.2 are a real cost to anyone doing authorised vulnerability work, and no amount of pro mode buys back a refusal.
Until then, the useful mental model is the boring one. GPT-6.2 Astra is a model with a price list; pro is a dial on it that spends more of the same tokens; and the marketplace string that concatenates the two is a label, not a checkpoint you can look up.
Compared in this article2
Detected from this article · Benchmarks: Artificial Analysis · updated daily
