Is GPT-6.1 Sol better than GPT-6 Astra?

For most work it is close and much cheaper. On LMArena, Sol ranks higher for chat, writing, roleplay, reasoning and vision, and Astra ranks higher for coding and agent tasks. OpenAI calls Sol "near-Astra performance at a lower cost" and keeps Astra for the hardest end-to-end work.

How much cheaper is GPT-6.1 Sol than Astra?

5x cheaper for input and output and 10x for cached input. On the API, Sol is $2 in, $0.10 cached and $10 out per 1M tokens; Astra is $10, $1 and $50.

Why does GPT-6 Astra use up my Codex limit so fast?

OpenAI estimates 5 to 45 Astra messages per 5 hours on Plus vs 15 to 160 on GPT-6.1 Sol, about a third as many. Astra also costs 5x the Codex credits per token (10x on cached input), and higher effort, Fast or Ultrafast mode use limits up faster still.

Can I use GPT-6 Astra on ChatGPT Plus in Codex?

Yes. OpenAI's Codex pricing page gives Plus an estimated 5 to 45 local Astra messages per 5 hours. On Free and Go, OpenAI points you to GPT-6 Luna instead.

How do I switch Codex to GPT-6.1 Sol?

Use /model in the CLI, launch with codex --model gpt-6.1-sol, or set model = "gpt-6.1-sol" in ~/.codex/config.toml to make it the default.

Is GPT-6.1 Sol available in regular ChatGPT chat?

Not per OpenAI's model docs: GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna are available in ChatGPT Work and Codex, not in Chat.

What does a coding day cost on each?

For 20M input tokens (16M cached) and 1M output, about $106 on Astra vs $19.60 on GPT-6.1 Sol at API list prices, or 2,650 vs 490 Codex credits.

GPT-6 Astra vs GPT-6.1 Sol: cost, Codex limits, which to use

API prices from OpenAI's pricing page, checked 2026-10-20. Codex credits and message estimates from OpenAI's Codex pricing page, checked 2026-10-20. Message counts are OpenAI's ranges, not fixed limits; your usage dashboard shows the real ones.

	GPT-6 Astra	GPT-6.1 Sol
API input per 1M	$10	$2
API cached input per 1M	$1	$0.10
API output per 1M	$50	$10
Over 272K-token prompts, in / out	$20 / $75	$4 / $15
Context window	1.05M	1.05M
Codex credits per 1M input	250	50
Codex credits per 1M cached input	25	2.5
Codex credits per 1M output	1,250	250
Codex Plus messages per 5 hours (estimate)	5 to 45	15 to 160
Ultrafast use of included usage	8x	8x

How much cheaper is GPT-6.1 Sol?

On the API, GPT-6.1 Sol is a fifth of Astra's price for input and output, and a tenth for cached input. Codex credits follow the same ratios: 250 vs 50 credits per 1M input tokens, 25 vs 2.5 cached, and 1,250 vs 250 output.

Example: a busy coding day of 20M input tokens, 16M of them cached, plus 1M output tokens. On the API that is about $106 on Astra and $19.60 on GPT-6.1 Sol. Through Codex credits it is about 2,650 credits on Astra and 490 on Sol, so the same work costs roughly 5.4x as much on Astra. Long agent sessions resend a lot of cached context, which is where Sol's 10x cheaper cached rate adds up.

Which is better?

LMArena task	GPT-6 Astra	GPT-6.1 Sol
Coding (Code Arena)	#2 of 23	#4 of 23
Chat (Text Arena)	#13 of 25	#10 of 25
Creative writing (Text Arena creative writing)	#14 of 25	#12 of 25
Roleplay and characters (Text Arena multi-turn)	#13 of 25	#11 of 25
Reasoning and math (Text Arena math)	#14 of 24	#12 of 24
Vision (Vision Arena)	#10 of 22	#2 of 22
Long documents (Document Arena)	#6 of 9	not ranked
Cheap bulk and agents (Agent Arena)	#2 of 23	#4 of 23

Rank among the models we track, from LMArena (CC BY 4.0), checked 2026-10-20. Both are scored at their best listed reasoning setting. Bold marks the higher rank.

OpenAI's own guidance says GPT-6.1 Sol "offers near-Astra performance at a lower cost" and recommends it for complex coding and agentic work when your account has it. It suggests Astra "for the hardest end-to-end work": complete workflows across code, apps and research that need sustained reasoning and judgment, where asking focused questions and holding the original constraints matters.

Why Astra eats your Codex limit

OpenAI's own Plus estimate drops from 15 to 160 messages per 5 hours on GPT-6.1 Sol to 5 to 45 on Astra, so you get roughly a third as much work per window. OpenAI does not publish exactly how included usage is counted and says credit prices alone don't determine it, but once you pay with credits Astra costs 5x as much per token (10x on cached input). Higher reasoning effort adds more tokens on top, and Fast mode uses included usage at 2.5x the Standard rate, Ultrafast at 8x. Weekly limits may also apply on Plus. Pro plans currently have no 5-hour limit, but weekly limits may apply.

How to switch Codex to GPT-6.1 Sol

In the CLI, type /model and pick gpt-6.1-sol, or start with codex --model gpt-6.1-sol.
To make it the default, put model = "gpt-6.1-sol" in ~/.codex/config.toml.
In the desktop app, use the model picker under the composer. OpenAI suggests starting Astra on Light and Sol on its default effort, then raising effort only when a task needs more planning.

Which plans include each

ChatGPT lists GPT-6 Astra and GPT-6.1 Sol on Plus, Pro 100, Pro 200 and Pro 500. Note that OpenAI says GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna run in ChatGPT Work and Codex, not in regular Chat. In Codex, OpenAI points Free and Go to GPT-6 Luna, not Astra or Sol; Plus ($20) and Business carry the per-model estimates above. GitHub Copilot lists both on Pro+ and Max. The full list is in which AI plan includes which model.

When Astra is worth it

Long, multi-step jobs where a wrong turn costs more than the tokens, like a migration plan or a hard bug that Sol keeps missing.
Work where you want the model to ask focused questions and keep your constraints in view.
Coding and agent tasks, where Astra ranks a little higher in the LMArena data above.

For day to day edits, reviews, refactors and chat, GPT-6.1 Sol gets you close for a fifth of the cost. To weigh other models too, open the model compare builder with these two loaded.

GPT-6 Astra vs GPT-6.1 Sol