GPT-6.1 Sol Pricing & Context Window

OpenAI's DevDay 2026 keynote on September 29 introduced GPT-6.1 Sol, a new model that lands one rung below GPT-6 Astra in capability but charges one-fifth of Astra's standard API prices — $2 per million input tokens and $10 per million output tokens. It is available to every paid tier (API, Plus, Pro, Business, Enterprise, and Edu) from day one, and GitHub Copilot added it to its model picker the same day. The pitch in OpenAI's own words: near-Astra performance for complex coding, computer use, and professional work.

If you came here comparing it against Claude Sonnet 5.5, the short version is uncomfortable for Anthropic: GPT-6.1 Sol costs exactly what Sonnet 5.5 costs — $2/$10 per million tokens — while shipping a slightly larger context window and OpenAI's full agentic toolchain. The rest of this post is the field guide: verified pricing down to cache and long-context rates, the Sonnet 5.5 and Astra comparisons, and where the model actually works today.

Quick answer

• Price: $2 input / $10 output per 1M tokens (standard) — exactly 1/5 of GPT-6 Astra. Cached input is $0.10, cache writes $2.50.
• Context: 1,050,000-token window (922K max input, 128K max output). Prompts over 272K input tokens bill the whole request at 2× input and 1.5× output rates.
• vs Sonnet 5.5: identical headline pricing ($2/$10); Sol halves cache reads ($0.10 vs $0.20), Sonnet 5.5 avoids Sol's above-272K long-context surcharge.
• vs GPT-6 Astra: Astra stays the flagship at $10/$50. Sol is the cost-rationalized pick when Astra's ceiling isn't needed.
• Availability: all paid ChatGPT plans and the API now; GitHub Copilot since September 29 (rolling out to Pro+, Max, Business, and Enterprise).
• Ultrafast mode: Astra only for now ($60/$300) — Sol Ultrafast is listed as "coming soon."

What GPT-6.1 Sol actually is

The GPT-6 family now has a clear ladder, straight from OpenAI's model catalog: Astra for the most demanding work, GPT-6.1 Sol to balance intelligence and cost, and Luna ($0.10/$0.50) for high-volume, cost-sensitive workloads. Confusingly, there is also an older GPT-6 Sol (no ".1") that shipped September 22 — GPT-6.1 Sol is the DevDay upgrade of that model, and the catalog now lists both at the same $2/$10 headline price. The visible upgrade is cache economics: GPT-6 Sol reads cached input at $0.20 per million, and the .1 halved that to $0.10.

Official model details worth knowing before you wire it up:

Two integration notes from the docs: tool calling requires the Responses API (Chat Completions works without tools), and none/minimal reasoning efforts are not supported — the floor is low. If you have used Claude Code's /effort slider, the low-to-max naming will feel familiar; here the model decides per step how much to think, and medium is the default.

GPT-6.1 Sol pricing: every rate on the card

OpenAI's announcement line — "one-fifth of Astra's standard API input and output token prices" — checks out exactly against the pricing page:

Details the table can't show:

• Cache writes bill at 1.25× the input rate ($2.50 per million) — relevant for long system prompts you rewrite often.
• The long-context surcharge applies to the full request, not just the overflow: a 300K-token prompt bills everything at 2× input / 1.5× output. At 922K max input, that ceiling matters for whole-repo prompts.
• Fast mode is 2× standard across the board, and Batch/Flex takes 50% off standard — the cheapest way to run Sol is $1/$5 in Batch.
• Ultrafast — the new DevDay speed tier that pushes up to 300 tokens per second in Codex — is currently priced only for Astra. Sol Ultrafast is "coming soon," so today's fast-path money quote is Fast mode at $4/$20.

One comparison the pricing table invites: Sol's cached input reads ($0.10) undercut even its own Batch input rate, which makes prompt caching unusually rewarding on this model. Agentic workflows that reuse a large system prompt between turns will feel this.

GPT-6.1 Sol vs Claude Sonnet 5.5

This is the matchup most teams will actually price, because Sonnet 5.5 ships as Claude Code's default Sonnet model since v2.1.284 and is generally available in GitHub Copilot since September 28:

Headline pricing is identical — OpenAI priced 6.1 Sol directly at Sonnet 5.5's number — so the real differences live in the fine print, and they pull in opposite directions:

• Caching favors Sol. Cached input reads cost $0.10 per million on Sol (5% of base) versus $0.20 on Sonnet 5.5 (10% of base). Agent loops that replay a large system prompt every turn — which is most coding agents — will run meaningfully cheaper on Sol.
• Long documents favor Sonnet 5.5. Anthropic bills Claude 4.6-and-later models at flat standard pricing across the entire 1M window — a 900K-token request costs the same per token as a 9K one. Sol switches your whole request to 2× input / 1.5× output above 272K input tokens, so whole-repo prompts are the one place Sonnet 5.5 is clearly cheaper.
• Max output is a wash at 128K tokens each, and the context windows differ by a rounding error (1.05M vs 1M).

Which model is better at your coding tasks is an empirical question — OpenAI's own guidance is to "compare it with Astra on your tasks." But the price excuse for staying in one ecosystem is now gone. If your workflow is Claude Code, our Claude Code effort and ultracode guide covers how Sonnet 5.5's effort defaults work; if you run Copilot, the model appears in the picker as it rolls out.

GPT-6.1 Sol vs GPT-6 Astra

Astra stays the flagship, and nothing in the DevDay announcements changed its pricing: $10/$50 standard (cached $1.00, long context $20/$75, Ultrafast $60/$300). Sol exists because most requests don't need Astra's ceiling:

• 5× cheaper at the headline rate, with the same 1.05M-token context window and the same tool-calling surface (Responses API, computer use via the Agents API).
• OpenAI's positioning is explicit: Sol delivers "near-Astra performance for complex coding, computer use, and professional work." The official model page tells you how to treat it — benchmark the two on your own tasks and pay 5× only where the gap shows.
• Astra Ultrafast exists today; Sol Ultrafast doesn't. If you need the 300-token-per-second tier right now, Astra is the only option at $60/$300.
• Practical read: route your default traffic to Sol, keep Astra for the hardest agentic runs, and you've reproduced the Astra/Luna split from September with a much smaller quality cliff in the middle.

Where you can use GPT-6.1 Sol today

• API: live on the Responses and Chat Completions endpoints, with Batch support; US and EU data residency (Fast mode excluded in EU).
• Codex CLI and Codex cloud: available in the model picker; the DevDay "Codex in the cloud" refresh made it usable from any device. Our Codex CLI tutorial covers the picker and the /model flow, and the model picker guide shows how to pin a default.
• ChatGPT: all paid plans (Plus, Pro, Business, Enterprise, Edu).
• GitHub Copilot: rolling out since September 29 to Pro+, Max, Business, and Enterprise across VS Code, the Copilot CLI, the coding agent, JetBrains, and Xcode — billed at provider list pricing under usage-based billing, and org admins can manage access via model policy.

The budget alternative this site exists for

$2/$10 is a mainstream frontier price, but it is still roughly an order of magnitude above DeepSeek's rates — the DeepSeek API starts around $0.15 per million input tokens off-peak, and Z.AI's coding plans go lower still (we track the current Z.AI discount separately). If your bottleneck is tokens-per-dollar on routine coding work rather than peak reasoning, the same agentic workflows run on those models today — the wiring guide for Codex with a near-free backend is in our Codex usage dashboard walkthrough.

FAQ

How much does GPT-6.1 Sol cost?

$2 per million input tokens and $10 per million output tokens at standard rates. Cached input is $0.10, cache writes are $2.50, Batch/Flex runs at $1/$5, and Fast mode doubles to $4/$20. Prompts above 272K input tokens bill the entire request at 2× input and 1.5× output rates. That is exactly one-fifth of GPT-6 Astra's standard pricing.

Is GPT-6.1 Sol cheaper than Claude Sonnet 5.5?

Not at the sticker level — the headline rates are identical at $2/$10 per million tokens. Sol's edge on the bill is a $0.10 cached-input rate (Sonnet 5.5 charges $0.20), which adds up in cache-heavy agent loops. Sonnet 5.5's edge is flat standard pricing across its full 1M window, while Sol applies 2× input / 1.5× output rates to entire requests above 272K input tokens. Real cost differences come from your cache-hit ratio and prompt sizes, not the sticker price.

GPT-6.1 Sol vs GPT-6 Astra — which should I use?

Default to Sol and escalate to Astra when a task actually needs the flagship's ceiling. They share the same context window and tool surface; Astra costs 5× as much at standard rates and is the only model with the Ultrafast speed tier today. OpenAI's own docs describe Sol as "near-Astra performance for complex work at a lower cost" and recommend comparing both on your workload.

Is GPT-6.1 Sol available on ChatGPT Free?

No — it shipped to paid plans: Plus, Pro, Business, Enterprise, and Edu, plus the API. Free and Go users remain on Luna-class models. In GitHub Copilot it is rolling out to paid Copilot tiers since September 29.

What is the GPT-6.1 Sol context window and input limit?

1,050,000 tokens of context, with 922,000 as the max input and 128,000 as the max output. Requests above 272K input tokens switch the whole request to long-context pricing (2× input, 1.5× output), which is the number to watch before you paste an entire repository.

Does GPT-6.1 Sol support Ultrafast mode?

Not yet. Ultrafast — up to 300 output tokens per second in Codex, 6× via API — is currently priced only for GPT-6 Astra at $60/$300 per million tokens. OpenAI lists Sol Ultrafast as "coming soon"; until then Sol tops out at Fast mode ($4/$20, 2× standard).