What Is GPT-6.1 Sol? API Pricing, GPT-6 Astra vs GPT-6 Sol, and How to Pay When Your Card Is Declined (2026)
GPT-6.1 Sol launched on DevDay 2026. Every line of the official price sheet, how to choose between it, GPT-6 Astra and GPT-6 Sol using only OpenAI's own (qualified) results, three worked cost examples, and two practical routes if your local card is declined.
1. What GPT-6.1 Sol is, in one paragraph
GPT-6.1 Sol is the upgrade to GPT-6 Sol that OpenAI released on September 29, 2026, the day of DevDay 2026. OpenAI's launch post says it is especially strong at agentic coding, computer use and professional work, brings Sol closer to GPT-6 Astra on hard tasks, and costs one-fifth of Astra's standard input and output token prices. Developers call it through the OpenAI API as gpt-6.1-sol. On the ChatGPT side, OpenAI says Plus, Pro, Business, Enterprise and Edu users can use it in ChatGPT Work and Codex, and notes that it is not yet available in chat. The official model page lists a 1,050,000-token context window with a 922,000-token maximum input and 128,000 maximum output tokens, an April 30, 2026 knowledge cutoff, text and image input, text output and reasoning token support. The Responses API, Chat Completions and Batch are supported, but tool calling requires the Responses API; Chat Completions only accepts requests without tools. It was one of the DevDay 2026 announcements; the main announcements are summarised at /hub/openai-devday-2026.
2. Reading the official price sheet line by line
Standard prices per million tokens on the model page: input $2, cached input $0.10, cache write $2.50, output $10. Read these rules alongside them. First, cached reads are 5% of the normal input rate. OpenAI's prompt caching guide says that for GPT-5.6 and later, most models read cache at 10% of the input rate and GPT-6.1 Sol at 5%, so 6.1 Sol's cached input is half of GPT-6 Sol's ($0.10 vs $0.20). Second, cache writes are billed at 1.25x the uncached input rate, so the first request that writes a prefix costs slightly more and the savings grow with each reuse. Third, long prompts double: once a single request exceeds 272K input tokens, the whole request is billed at 2x input and cache rates and 1.5x output. Fourth, Batch and Flex are 50% cheaper than Standard, Fast mode is 2x Standard, and regional processing adds 10% where available. Fifth, the reasoning guide states that reasoning tokens are invisible in the API but billed as output tokens, so higher reasoning effort usually means a larger output bill. Prices are as listed on developers.openai.com in early October 2026; always check the live page.
3. GPT-6.1 Sol vs GPT-6 Astra: OpenAI's own numbers, with their conditions
One caveat first: every number below comes from OpenAI's launch post and was measured on specific benchmarks at specific reasoning settings. Your workload may behave differently. Coding: on DeepSWE v1.1, OpenAI says GPT-6.1 Sol performs comparably to GPT-6 Astra at about one-fifth of the cost. Computer use: on OSWorld 2.0 at the highest reasoning effort, the gap to Astra narrows to within 2.1 percentage points at about one-seventh of the per-task cost. Science: on Terminal-Bench Science 0.1 at the highest reasoning setting, the average per-task cost was $5.47 for GPT-6.1 Sol and $23.80 for Astra, yet OpenAI also states that Astra still ranks first among the models tested, with a 68.1% score, and should be used for the most challenging scientific research tasks. Factuality: at extra-high reasoning, the factual error rate was 4.1% for GPT-6.1 Sol and 4.0% for Astra, on prompts OpenAI says were deliberately selected to be difficult and are not representative of typical use. OpenAI's model selection guide is direct about the trade-off: if cost and latency are not a concern, you can default to Astra; for complex projects where cost matters, consider GPT-6.1 Sol and compare it with Astra on the same task, then keep the lightest setting that meets your quality bar. The guide suggests 6.1 Sol at medium effort for complex technical work and coordinated deliverables you expect to revise, and at extra-high effort for polished deliverables, connected visual systems and decisions built from conflicting evidence.
4. Already on GPT-6 Sol? Whether to move to 6.1
On price, both list at $2 input and $10 output, 6.1 Sol's cached reads are cheaper ($0.10 vs $0.20), and its knowledge cutoff is ten days newer (April 30 vs April 20). On OpenAI's evaluations, 6.1 Sol scores 6.4 points above GPT-6 Sol's best result on DeepSWE v1.1 and 7 points higher on OSWorld 2.0 at the highest reasoning effort. On the same deliberately difficult prompts at extra-high reasoning, its factual error rate was 4.1% vs 4.5%, which OpenAI notes is not representative of typical use. Three things change when you migrate, per the GPT-6 guide and the 6.1 Sol model page. One: 6.1 Sol does not support the none or minimal reasoning efforts (GPT-6 Sol supports none); OpenAI suggests starting at low and comparing. Two: tool calling must go through the Responses API. Three: when reasoning effort is not none, remove temperature, top_p and top_logprobs; on Chat Completions also remove logprobs, and on Responses remove message.output_text.logprobs from include. If you are coming straight from GPT-5.5 or earlier, there is one more change: replace prompt_cache_retention with prompt_cache_options.ttl (GPT-6 guide); OpenAI's prompt caching guide says its only supported value is 30m. If your workload is mostly small, high-volume tasks, look at GPT-6 Luna first: OpenAI calls it the most efficient model, and its standard input and output prices are one-twentieth of 6.1 Sol's.
5. Three worked cost examples at the official rates
All figures use official standard prices with inputs of up to 272K per request unless stated, and only illustrate the arithmetic; your real usage is whatever the usage field in each response says. Example 1: a coding request with 30,000 input tokens and 8,000 output tokens (reasoning included), no caching. GPT-6.1 Sol: 30,000 x $2 / 1M = $0.06 plus 8,000 x $10 / 1M = $0.08, total $0.14. The same token counts cost $0.30 + $0.40 = $0.70 on GPT-6 Astra and $0.003 + $0.004 = $0.007 on GPT-6 Luna. Over 1,000 runs that is about $140, $700 and $7. Note that different models use different numbers of tokens for the same task, so compare real bills on the same batch of tasks rather than assuming equal token counts. Example 2: a fixed 50,000-token prefix (system prompt plus code context) reused 20 times within 30 minutes, each call adding 2,000 new input tokens and producing 3,000 output tokens. Without caching: 1.04M input tokens is about $2.08, 60,000 output tokens is about $0.60, total about $2.68. With an explicit cache breakpoint on the prefix only, and assuming the last 19 calls all hit: one write of 50,000 x $2.50 / 1M = $0.125, 950,000 cached reads at about $0.095, 40,000 fresh input tokens at about $0.08, and output still $0.60, for a total of about $0.90. Example 3: one request with 300,000 input tokens and 10,000 output tokens. It crosses 272K, so the whole request bills input at $4 and output at $15: $1.20 + $0.15 = $1.35. Trim the input to 270,000 tokens and it bills at standard rates: $0.54 + $0.10 = $0.64. For long documents and large codebases, trimming before you send is often the simplest saving.
6. Minimal call against the official OpenAI API
Per the official docs, set model to gpt-6.1-sol in a Responses API request. reasoning.effort accepts low, medium (the default), high, xhigh and max. The example below uses your own OpenAI API key and sends the request to OpenAI's endpoint, so it is billed to your own OpenAI account.
curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6.1-sol",
"reasoning": { "effort": "medium" },
"input": "Summarize the tradeoffs of using a cheaper model for code review."
}'7. Route one: the official OpenAI API, paid with a US-BIN virtual card
This route gives you the genuine model, with the account, the key and the bill all in your own hands. Do every step yourself on the official pages and never hand your account to anyone. Step 1: create your own OpenAI account at platform.openai.com. Step 2: sign up for cocodot, top up your wallet with Alipay, open a US-BIN Visa or Mastercard virtual card in the dashboard, and move money from the wallet onto the card. Wallet balance and card balance are separate; money not moved onto the card cannot be charged. Step 3: add the card on OpenAI's Billing page and fund your API balance there. API top-ups and usage charges are OpenAI's own billing, so how much to add, how usage is deducted and what happens when the balance runs out all follow what your OpenAI Billing page shows. Step 4: create an API key in the official console and call the model as in the previous section. cocodot's US-BIN virtual cards have had successful charges for OpenAI API top-ups, but risk rules on both the issuing side and the merchant side change over time, so no card can guarantee that every charge goes through. One more thing: OpenAI publishes an official list of supported countries and territories (developers.openai.com/api/docs/supported-countries); check it yourself before signing up and using the service, and make sure your use complies with OpenAI's terms and the laws where you are.
8. Why payments to OpenAI get declined
The card's issuing country: some platforms use the first digits of the card number (the BIN) to determine where a card was issued, and cards from some regions may be declined on cross-border online charges even when they carry the Visa or Mastercard logo; what needs to change is the BIN, not the brand. Billing address: enter the complete address you actually use, and keep it the same for the same account; never use an address copied from the internet, someone else's, or a made-up one. Insufficient card balance: a virtual card is funded before it is spent, so leave a margin when you load it. Rapid retries: after two failures in a row, stop and check the balance, the address and the card status in that order, because repeated retries may trigger risk controls. The merchant name on your card statement starts with OPENAI; reconcile by merchant name, amount and date. A fuller troubleshooting order is at /hub/openai-api.
9. Route two: an OpenAI-compatible endpoint you can top up with Alipay
The boundary first: cocodot's API relay does not carry the GPT-6 family today (no GPT-6 Astra, GPT-6 Sol, GPT-6.1 Sol or GPT-6 Luna), and it does not offer the Decisions API. If you specifically need gpt-6.1-sol, use route one. If what you need is a strong model you can call from code without opening an overseas account, paid as you go with Alipay, cocodot offers GPT-5.6 Sol / Terra / Luna, GPT-5.5, Claude (Fable 5, Opus 5, Sonnet 5 and others), Gemini, DeepSeek, Qwen, GLM, Kimi and more in OpenAI-compatible format. Point base_url at https://cocodot.co/api/ai/v1 and use your own key. Live per-model prices are on /pricing; how to pick between the three GPT-5.6 tiers is at /hub/gpt-56-sol-terra-luna-api, a comparison with Claude is at /hub/opus-5-vs-gpt-56-sol, and the three access routes for Claude and GPT are compared at /hub/claude-gpt-api-china-guide.
from openai import OpenAI
client = OpenAI(
base_url="https://cocodot.co/api/ai/v1",
api_key="YOUR_COCODOT_KEY",
)
resp = client.chat.completions.create(
model="gpt-5.6-sol", # gpt-6.1-sol is not available on cocodot; see /pricing for models
messages=[{"role": "user", "content": "Explain prompt caching in three sentences."}],
)
print(resp.choices[0].message.content)10. No code, just ChatGPT or Codex
According to OpenAI, GPT-6.1 Sol is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and is not yet available in chat. The Codex models page in OpenAI's ChatGPT help documentation (learn.chatgpt.com) adds that Enterprise and Edu keep GPT-6.1 Sol off by default until an administrator enables it, and that Free and Go are not included at launch. That route requires a ChatGPT plan at one of those tiers; plans and prices follow OpenAI's official pages; setup steps are at /hub/chatgpt-plus and /hub/chatgpt-pro. cocodot's US-BIN virtual cards have had successful charges for ChatGPT subscriptions and have also seen declines, so no card can guarantee every charge. Subscriptions renew automatically, so move funds onto the card before the renewal date.
11. Sources and limits of this article
Every price, specification and benchmark statement about GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol and GPT-6 Luna in this article comes from OpenAI's GPT-6.1 Sol launch post and DevDay 2026 recap on openai.com, and from the model pages, GPT-6 guide, model selection guide, prompt caching guide and reasoning guide on developers.openai.com, as read in early October 2026. Prices and availability follow OpenAI's latest pages. Around DevDay, a number of speed figures and insider claims about other new models and APIs circulated online; we could not find them on any official page, so none are cited here. What cocodot offers on this topic is two things: US-BIN virtual cards for paying your own OpenAI account (see /card), and an Alipay-funded, OpenAI-compatible API relay that does not currently include the GPT-6 family (see /pricing). Pick whichever matches what you need.
GPT-6 family official API standard prices (USD per 1M tokens, input up to 272K per request; source: developers.openai.com model pages, checked 2026-10-04)
| Model (API name) | Input | Cached input | Cache write | Output | OpenAI's positioning |
|---|---|---|---|---|---|
| GPT-6 Astra (gpt-6-astra) | $10 | $1 | $12.50 | $50 | Most capable; for the most demanding work: complex reasoning, coding, computer use, research and document creation |
| GPT-6.1 Sol (gpt-6.1-sol) | $2 | $0.10 | $2.50 | $10 | Near-Astra performance at a lower cost for complex coding, computer use and professional work |
| GPT-6 Sol (gpt-6-sol) | $2 | $0.20 | $2.50 | $10 | Built for complex coding and agentic workflows; its model page now points to 6.1 Sol |
| GPT-6 Luna (gpt-6-luna) | $0.10 | $0.01 | $0.125 | $0.50 | Most efficient; for focused, high-volume tasks |