What does each flagship AI model cost and hold?
Any AI model comparison starts with price, because that is where the flagships split, and the answers cluster in an odd place. Four models from four vendors list at exactly $2 per million input tokens and $10 per million output tokens on 15 November 2026: Claude Sonnet 5.5, GPT-6.1 Sol, Gemini 4 Carbon and Kimi K3.1 Pro. Google's price is a promotion that ends November 30.

The chart uses one standard answer, which is 1,000 tokens in and 500 out, priced at the list rate of each vendor. The table below has the full per-million prices, and it is the one you can check against the vendors. Every cell comes from the vendor's own pricing or model page, read on 15 November 2026.
| Model | Context window (API) | Input, per 1M tokens | Output, per 1M tokens |
|---|---|---|---|
| OpenAI GPT-6 Astra | 1.05M | $10 | $50 |
| OpenAI GPT-6.1 Sol | 1.05M | $2 | $10 |
| Anthropic Claude Fable 5.2 | 1M | $8 | $40 |
| Anthropic Claude Opus 5.5 | 1M | $4 | $20 |
| Anthropic Claude Sonnet 5.5 | 1M | $2 | $10 |
| Google Gemini 4 Carbon | 1M | $2.00 until Nov 30, then $4.00 | $10.00 until Nov 30, then $20.00 |
| Google Gemini 4 Argon | 1M | $2.00 until Nov 30, then $4.00 | $10.00 until Nov 30, then $20.00 |
| Google Gemini 4 Hydrogen | 1M | $1.50 | $7.50 |
| xAI Grok 4.7 | 500K | $2.00 | $6.00 |
| DeepSeek V4 Pro | 1M | $1.32 peak, $0.66 off-peak | $3.96 peak, $1.98 off-peak |
| Moonshot Kimi K3.1 Pro | 1,048,576 | $2.00 | $10.00 |
Two vendors price by the clock or the calendar. DeepSeek halves its rates outside 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, and Gemini 4 Carbon and Argon double on December 1. Treat any price here as a snapshot, not a quote. If you only compare one thing, compare the Sonnet, Sol and Carbon rows, because they carry the same price and you can see what it buys from each vendor.
Is GPT-6 out, and which models count as flagships?
Yes. OpenAI's API changelog shows GPT-6 Astra released on September 3, 2026, GPT-6 Sol and Luna on September 22, GPT-6.1 Sol on September 29 and GPT-6.1 Terra on October 26. The October 6 entry people cite is the Decisions API beta, not the model launch.
Each vendor names its own starting point. OpenAI's model page says "If you're not sure where to start, use GPT-6 Astra, our flagship model for complex reasoning and coding." Anthropic's says "If you're unsure which model to use, start with Claude Opus 5.5 for most workloads." Astra output costs 2.5 times Opus 5.5 output.
Google lists three Gemini 4 models, Carbon, Argon and Hydrogen. xAI's docs name Grok 4.7 as the model for "code and everything else", released September 21, and no Grok 5 appears in its release notes. DeepSeek sells deepseek-v4-pro and deepseek-flash. Moonshot sells Kimi K3.1 Pro. Anthropic also lists Claude Mythos 5.2, but only verified organizations can use it, so it stays out of the table.
Which consumer plan gets which model?
The cheapest way into a top model is not the same on every app. As of November 2026, ChatGPT Plus at $20 a month lists GPT-6 Astra, GPT-6.1 Sol and GPT-6 Sol, while Claude's equivalent top tier, Fable, is paid for with usage credits on Pro and included on Max at $100 a month.
| Model | Cheapest plan that reaches it | Plan price | Source row |
|---|---|---|---|
| GPT-6 Astra | ChatGPT Plus (Pro "Expanded") | $20 a month | chatgpt.com/pricing, Models |
| GPT-6.1 Sol | ChatGPT Plus | $20 a month | chatgpt.com/pricing, Models |
| GPT-6 Luna | ChatGPT Free | $0 | chatgpt.com/pricing, Models |
| Claude Fable | Claude Pro with usage credits, or Max with up to half the weekly limit | $20 a month ($17 on annual billing) or $100 | claude.com/pricing, Fable row |
| Claude Opus | Claude Pro | $20 a month ($17 on annual billing) | claude.com/pricing, Opus row |
| Claude Sonnet | Claude Free | $0 | claude.com/pricing, Sonnet row |
| Gemini 4 Hydrogen | Google AI Plus | $4.99 a month | one.google.com, AI plans |
| Gemini 4 Argon | Google AI Pro | $19.99 a month | one.google.com, AI plans |
| Gemini 4 Carbon | Google AI Ultra | From $99.99 a month | one.google.com, AI plans |
| Grok (latest models) | Grok Premium | $30 a month | x.ai/pricing, Premium list |
| DeepSeek V4 Pro | API only, pay as you go | Per token | api-docs.deepseek.com |
| Kimi K3.1 Pro | API only, pay as you go | Per token | platform.kimi.ai |
Read the Google rows twice. Plus carries the cheapest Gemini 4 model and Pro adds Argon. Only the Ultra tier, at five times the usage of Pro, is listed with both Argon and Carbon. xAI's page says that Premium includes "Latest Grok models" and it doesn't name Grok 4.7, so the table says only what the page says. Kimi's API page says it has no subscription plan and calls Kimi membership a separate product, which this page didn't price.
What if the model you want sits on a plan you don't have?
Three apps cost money before any of them gives you a top model. ChatGPT Plus is $20, Claude Pro is $20 and Google AI Pro is $19.99 a month, a total of $59.99 a month for the Astra, Opus and Argon tiers.
Whizi Pro is one app that carries Claude, GPT and Gemini models side by side for $29.99 a month, and its first week is $1.99. You choose the model per chat, so one prompt can run on two vendors and you judge the answers. The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, with no card.
Whizi loses on two points. The newest tier from a vendor can reach the Whizi picker later than it reaches the vendor's own app, and the features that a vendor builds only for its own product stay there. Check the model picker for what is live before you pay.
Is the app context window the same as the API's?
No. The API figures in the first table are ceilings for developers. ChatGPT's pricing page lists a reasoning context window of 256K on Plus and 400K on Pro, and 54K on Plus for the instant models, against 1.05M for GPT-6 Astra in the API.
Claude's pricing page says "Up to 1M" on every plan and adds that it varies by model. Google lists an expanded 1 million token window starting on AI Pro. So a 1M figure in a comparison chart tells you about the model, and the plan decides how much of it you get.
If your job is one long contract, you should check the limit inside the app that you pay for before you trust the 1M figure. Context window comparison walks through how many pages each size holds.
Why this table prints no benchmark scores
Benchmark numbers belong on the benchmark's own site with a date, and this page was built from vendor pricing and model pages only. A score that comes from a launch post measures the setup of that vendor, and the same test name can hide different conditions on two charts.
For a pick by task, the best AI model for every task routes writing, code, math and long documents to named models. For a method you can run in ten minutes on your own prompt, use how to choose an AI model. Your own three prompts beat any chart, this one included.
Which flagship fits your budget and your plan?
It depends on what you already pay for and what you do all day. These are the cases where the table gives a clear answer.
Pick GPT-6 Astra if you want the model that OpenAI calls its flagship and you already pay $20 for ChatGPT Plus. It has the highest list price in the table, $10 in and $50 out, so it is the one to ration in the API.
Pick Claude Opus 5.5 if you write or code all day on Claude, since Anthropic itself says to start there. At $4 and $20 it costs half of what Fable 5.2 does and a fifth less than Claude Opus 5, which the Anthropic models page lists at $5 and $25 among its older models.
Pick Gemini 4 Carbon through the API before December 1 if you want a top tier at $2 and $10, which is one fifth of the output price of Astra. It doubles after that.
Pick Grok 4.7 or DeepSeek V4 Pro when you send many requests. Grok lists $2 and $6 with a 500K window, and DeepSeek charges $0.66 and $1.98 when you call it off-peak.
Pick GPT-6.1 Sol or Claude Sonnet 5.5 if you want one price, $2 and $10, and a 1M window from either vendor.
- Check the vendor's own pricing page before you budget, because three rows in this table change on a stated date
- Compare the app's context limit, not only the API figure, if you work in ChatGPT or Claude directly
- Price one standard answer, 1,000 tokens in and 500 out, before you compare per-million rates
- Run your own three prompts through two models before you pick a default
Frequently asked questions
What is the best AI model right now?
No page that only reads vendor pricing can name one. As of November 2026 each vendor points to a different starting model: OpenAI to GPT-6 Astra and Anthropic to Claude Opus 5.5. Pick by task and price, then test two models on your own prompt.
Which AI model is cheapest per million tokens?
Among the eleven flagships here, DeepSeek V4 Pro is cheapest at $1.32 in and $3.96 out at peak, half that off-peak. Gemini 4 Hydrogen is $1.50 and $7.50, and Grok 4.7 is $2.00 and $6.00. GPT-6 Astra is the most expensive at $10 and $50.
Is GPT-6 released?
Yes. OpenAI's API changelog lists GPT-6 Astra on September 3, 2026, and its model page names Astra the flagship, with GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna below it. ChatGPT Plus lists Astra, Sol and Luna.
Do these models all have 1M-token context?
Almost. Astra and Sol list 1.05M tokens, the Claude models and Gemini 4 models 1M, DeepSeek V4 Pro 1M and Kimi K3.1 Pro 1,048,576. Grok 4.7 lists 500K. Apps cap it lower on some plans, as ChatGPT does.