Wain AI/Tech Blog

AI news and trends worldwide, updated nearly every day

OpenAI Releases GPT-6.2 Astra and Sol - Astra Is 20% Cheaper, and Migration Can Drop Reasoning Items

OpenAI Releases GPT-6.2 Astra and Sol - Astra Is 20% Cheaper, and Migration Can Drop Reasoning Items

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026. In the API, Astra costs $8 input and $40 output, 20% below GPT-6 Astra, while Sol keeps GPT-6.1 Sol's prices. Rewriting history mid-conversation now discards reasoning items, among other points to check before migrating.

On November 17, 2026 (US time), OpenAI announced GPT-6.2 Astra and GPT-6.2 Sol in its API changelog 1. Astra is aimed at hard reasoning and coding work, and Sol at day-to-day coding and work carried out by agents. Each of them runs on the Responses API and on Chat Completions, although tools can only be called through the Responses API.

The models reached ChatGPT Work and Codex 7 and GitHub Copilot 8 the same day. For API users, the questions are how the per-token prices moved and whether an existing integration can simply switch the model name.

Astra drops 20% in price; Sol stays put

Here are the official per-million-token prices next to the previous generation (Standard processing, short context) 2356.

ItemGPT-6 AstraGPT-6.2 AstraGPT-6.1 SolGPT-6.2 Sol
Input$10$8$2$2
Cache read$1$0.4$0.1$0.1
Cache write (30 min)$12.5$10$2.5$2.5
Output$50$40$10$10

Against GPT-6 Astra, GPT-6.2 Astra is 20% cheaper on input, output, and cache writes, and 60% cheaper on cache reads. Cache reads now cost 5% of the input price, the same ratio as GPT-6.1 Sol. For workloads that resend the same prefix again and again, the lower read price is likely to show up on the bill as well.

GPT-6.2 Sol, by contrast, matches GPT-6.1 Sol on all four items. An organization already on Sol pays the same Standard rates after switching.

Some terms apply to both models. Once a request’s input goes past 272K tokens, input and cache are billed at double and output at 1.5 times. Batch is half the Standard rate, and Fast is twice it 2.

A 12-hour cache write is priced at $20 for Astra and $5 for Sol 23. Anyone using the 12-hour cache added in October should estimate costs with these figures.

Ultrafast at 3x until the end of February 2027

For both GPT-6.2 models, Ultrafast is priced at three times Standard through February 28, 2027, and the multiplier becomes six on March 1, 2027, Pacific time 23. GPT-6.1 Sol’s Ultrafast is six times Standard 6, so the 3x rate for GPT-6.2 is time-limited. Budgets that assume Ultrafast need to allow for the higher multiplier from March.

The two models share the same specs: a 1,050,000-token context (922,000 maximum input), 128,000 maximum output tokens, text and image input, and text output 2. The knowledge cutoff is May 18, 2026 for Astra and June 1, 2026 for Sol 3.

What the migration guide asks you to check

The migration guide lists points to check when moving to GPT-6.2 4.

First, reasoning items. From GPT-6.2 on, the API throws reasoning items away once it notices that the conversation history was edited partway through (a prefix edit), a model downgrade, or a change of end user or organization. Once they are gone, answer quality can suffer, and the model may spend extra tokens redoing earlier work. The guide asks developers to find places where instructions or tool calls change mid-conversation, and to leave the prefix untouched even when settings change. An unchanged prefix also lets earlier input keep qualifying for the prompt cache, it says.

Second, parameters. Chat Completions requests need to drop temperature, top_p, logprobs, and top_logprobs; Responses requests need to drop temperature, top_p, and top_logprobs, and remove message.output_text.logprobs from include. The API refuses a reasoning effort of none 2. The guide suggests moving from none or minimal to low, and starting new integrations at medium.

Third, prompt caching settings. When coming from GPT-5.5 or earlier, prompt_cache_retention has to be replaced with prompt_cache_options.ttl ("30m"). Data residency covers the US and the EU for Standard, Fast, and Batch, but Ultrafast is US-only.

Output style differs too. According to the guide, Astra writes more concisely and directly by default, and the guide recommends revisiting instructions such as “be concise” in system prompts. It also notes that when intent is unclear, Astra may propose what to do next and hold off until the user confirms.

ChatGPT Work, Codex, and GitHub Copilot on the same day

In ChatGPT Work and Codex, both models can be selected from November 17. OpenAI says GPT-6.2 Sol gets more work done than GPT-6.1 Sol without moving to a different paid plan. Ultrafast is available with GPT-6.2 Sol on the Pro $200 plan, and with both Sol and Astra on the Pro $500 plan 7.

In GitHub Copilot, the models are generally available on the Pro+, Max, Business, and Enterprise plans. Usage is billed at the provider’s list price under usage-based billing, and the rollout is gradual. GitHub says that in its early testing, GPT-6.2 Sol surfaced close to double the genuine problems that GPT-6.1 Sol did when reviewing code 8.

What Copilot administrators should check is the default. New models switch on by themselves unless an admin has disabled the global default or blocked these models one by one 8. As with the general availability of GPT-6 Astra, organizations that restrict which models their members use should revisit the Copilot model policy.

What to weigh before switching

On price alone, Astra users get lower Standard rates just by switching, and Sol users see no change. The deciding factor is likely to be how much work the reasoning-item behavior requires. Integrations that swap system prompts or tool definitions partway through a conversation should be tested before the model is changed.

The most reliable way to see the difference in real work is to compare on your own workloads, with the migration guide’s caveats in mind.

Sources

  1. Changelog - OpenAI API official changelog (entry for November 17, 2026)
  2. GPT-6.2 Astra - OpenAI official documentation (pricing and specs, checked November 18, 2026)
  3. GPT-6.2 Sol - OpenAI official documentation (pricing and specs, checked November 18, 2026)
  4. Using GPT-6.2 - OpenAI official migration guide (checked November 18, 2026)
  5. GPT-6 Astra - OpenAI official documentation (comparison pricing, checked November 18, 2026)
  6. GPT-6.1 Sol - OpenAI official documentation (comparison pricing, checked November 18, 2026)
  7. Changelog - ChatGPT Work and Codex official changelog (entry for November 17, 2026)
  8. OpenAI’s GPT-6.2 Astra and GPT-6.2 Sol now available - GitHub official changelog (November 17, 2026)
  9. Introducing GPT-6.2 Astra and Sol - OpenAI official announcement (November 17, 2026)

We publish the latest AI news nearly every day.

Subscribe via RSS Get new posts the moment they go live.

Search other keywords →