DevTk.AI
GPT-6.2GPT-6.2 AstraGPT-6.2 SolOpenAI APIAPI PricingCodex

GPT-6.2 Astra and Sol: API Prices, What Changed, and Whether to Switch

OpenAI released GPT-6.2 Astra ($8/$40) and GPT-6.2 Sol ($2/$10) on November 17, 2026. Prices, cache and long-context rates, Ultrafast at 3x until February, Codex limits, and the cost math for moving off GPT-6 Astra, GPT-6.1 Sol and older models.

By Published 6 min read

Short answer (checked against OpenAI’s pricing page on November 18, 2026): OpenAI released two models on November 17, 2026. GPT-6.2 Astra costs $8 input / $40 output per 1M tokens, 20% less than GPT-6 Astra ($10 / $50), with cache reads 60% cheaper. GPT-6.2 Sol costs $2 / $10, the same as GPT-6.1 Sol. Both have a 1.05M-token context window and 128K max output. If you run Astra today, switching cuts the bill by about a fifth for the same token counts. If you run GPT-6.1 Sol, the token price doesn’t change, but Ultrafast mode costs half as much until February 28, 2027.

For the full GPT-6 family (Terra, Luna, older Sol) see the GPT-6 API pricing guide, and for every OpenAI model on one page see the OpenAI API pricing overview.

GPT-6.2 Prices (Standard, prompts up to 272K tokens)

USD per 1M tokens:

ModelInputCached inputCache write, 30 minCache write, 12 hOutput
GPT-6.2 Astra$8.00$0.40$10.00$20.00$40.00
GPT-6.2 Sol$2.00$0.10$2.50$5.00$10.00
GPT-6 Astra (previous)$10.00$1.00$12.50$25.00$50.00
GPT-6.1 Sol (previous)$2.00$0.10$2.50$5.00$10.00

OpenAI describes GPT-6.2 Astra as “our most capable model for your hardest projects” and GPT-6.2 Sol as “our go-to model for everyday work, from coding to powering agents at scale”. Both accept text and image input, return text, and support function calling and structured outputs. Maximum input is 922K tokens.

Long Prompts (above 272K input tokens)

When a prompt goes over 272K input tokens, the whole request is billed at the long-context rate, not just the extra tokens:

ModelInputCached inputCache write, 30 minOutput
GPT-6.2 Astra$16.00$0.80$20.00$60.00
GPT-6.2 Sol$4.00$0.20$5.00$15.00
GPT-6 Astra$20.00$2.00$25.00$75.00

A 300K-token prompt is billed at these rates from its first token. Keeping prompts under 272K, for example by trimming retrieved context, halves the input price.

Batch, Fast and Ultrafast

ModeMultiplierGPT-6.2 Astra (in / out)GPT-6.2 Sol (in / out)
Batch50% of Standard$4 / $20$1 / $5
Standard1x$8 / $40$2 / $10
Fast2x$16 / $80$4 / $20
Ultrafast, until 2027-02-283x$24 / $120$6 / $30
Ultrafast, from 2027-03-016x$48 / $240$12 / $60

The Ultrafast rate is the biggest short-term change. GPT-6 Astra and GPT-6.1 Sol charge 6x for Ultrafast ($60 / $300 and $12 / $60), while the GPT-6.2 models charge 3x until the end of February. If you pay for Ultrafast today, moving to GPT-6.2 halves that line of the bill for about three months.

Two other modifiers apply to all of these: the Responses API’s background completion windows (public beta) cost 30% less than Standard for a 2-hour window and 40% less for an 8-hour window (also offered on GPT-6.1 Terra and GPT-6 Luna, but not on GPT-6 Astra or GPT-6.1 Sol), and regional data-residency and FedRAMP endpoints add 10%.

Cost Math: Should You Switch?

Example workload: a coding agent that sends 20M input tokens a month, 70% of them cache hits, and receives 2M output tokens. Standard mode, prompts under 272K, cache-write charges left out for simplicity.

ModelUncached inputCached inputOutputMonthly total
GPT-6 Astra6M × $10 = $6014M × $1.00 = $142M × $50 = $100$174.00
GPT-6.2 Astra6M × $8 = $4814M × $0.40 = $5.602M × $40 = $80$133.60
GPT-6.1 Sol / GPT-6.2 Sol6M × $2 = $1214M × $0.10 = $1.402M × $10 = $20$33.40
  • From GPT-6 Astra: about 23% cheaper for the same tokens. Cached-heavy workloads save more because the cache-read price fell 60%.
  • From GPT-6.1 Sol: the same price in Standard mode. The reason to switch is that 6.2 Sol is the newer model, and Ultrafast is 3x instead of 6x until February 28, 2027.
  • From older models: GPT-5, GPT-5 mini and nano, o3 and o3-pro shut down on December 11, 2026. GPT-4.1, GPT-4.1 mini and gpt-4o-mini shut down on January 14, 2027. Anything still on those needs a new model soon, and GPT-6.2 Sol at $2 / $10 is the natural default.

Different models can use different numbers of tokens for the same task, so treat these as per-token comparisons and measure your own workload before committing.

GPT-6.2 in Codex

GPT-6.2 Astra and Sol are available in Codex on ChatGPT Plus and above. OpenAI’s estimates for local messages per 5 hours on Plus:

ModelMessages per 5 hours (Plus)
GPT-6.2 Astra6–50
GPT-6.2 Sol17–180
GPT-6.1 Sol15–160
GPT-6.1 Terra35–285
GPT-6 Luna350–3,000

Usage beyond the plan limits on GPT-6.2 models is charged at 3x until February 28, 2027 and 6x from March 1. See the AI coding tools cost hub for how Codex plans compare with Claude Code, Cursor and Copilot.

Which GPT Model Is the Cheapest One That Does the Job?

Start from the cheapest tier that can handle the task, and escalate only the requests that fail:

  • GPT-6 Luna ($0.10 / $0.50): OpenAI positions it for focused, repeatable, high-volume work such as classification, extraction and routing.
  • GPT-6.1 Terra ($1 / $5): half Sol’s price, aimed at visual reasoning, computer use and natural writing.
  • GPT-6.2 Sol ($2 / $10): the everyday default for coding and agents. New projects should start here rather than on GPT-6.1 Sol, since the price is the same.
  • GPT-6.2 Astra ($8 / $40): for the hardest end-to-end work. It replaces GPT-6 Astra for new work at a lower price.

To compare these against Claude, Gemini, DeepSeek and others for your own token volumes, use the AI pricing calculator.

Sources

Next step

Related Posts