AI API pricing this week: GPT-6.1 Terra launches at $1/$5, Kimi K3.1 Pro undercuts K3 by a third and Gemini 4 Carbon debuts at $2/$10
The week from October 23 to October 30, 2026 brought 13 verified changes to the price lists and terms of the main AI APIs, spread across OpenAI, Moonshot AI, Google and Anthropic. Three new price lists set the tone: OpenAI released GPT-6.1 Terra at $1.00 per million input tokens and $5.00 per million output tokens, half the rates of GPT-6.1 Sol; Moonshot AI launched Kimi K3.1 Pro at $2.00 input and $10.00 output, a third below the rates it lists for Kimi K3; and Google announced Gemini 4 Carbon with introductory pricing of $2 input and $10 output that later doubles to $4 and $20.
The rest of the week was about terms and limits rather than headline prices. Google switched on request storage by default for new paid projects that use the Gemini Generate Content API, Anthropic raised the maximum request size of the Claude Messages and Token Counting APIs from 32 MB to 128 MB, and OpenAI began giving eligible ChatGPT Pro and Business subscribers up to $500 a month in API credits. Further scheduled changes at Moonshot AI, DeepSeek and Google are listed separately because their notices carry no announcement date.
All prices below are in US dollars per one million tokens, with input and output listed separately and cached and batch prices given where the provider publishes them. Prices as listed on the official pricing pages and official documentation on October 30, 2026; where a figure appears only in an official announcement and not yet on a pricing page, the text says so.
··········
ALL API PRICE AND TERMS CHANGES THIS WEEK.
The table lists every change verified this week, one row per provider, model and priced item or rule. For new models there is no previous price, so the "Before" column shows the provider's closest existing model in the same family for reference.
........
Provider | Model | Item | Before | After | Effective date |
OpenAI | GPT-6.1 Terra | Input | New model (GPT-6.1 Sol: $2.00) | $1.00 | Oct 26 |
OpenAI | GPT-6.1 Terra | Cached input | New model (GPT-6.1 Sol: $0.10) | $0.05 | Oct 26 |
OpenAI | GPT-6.1 Terra | Output | New model (GPT-6.1 Sol: $10.00) | $5.00 | Oct 26 |
OpenAI | GPT-6.1 Terra | Batch input / output | New model (GPT-6.1 Sol: $1.00 / $5.00) | $0.50 / $2.50 | Oct 26 |
OpenAI | API credits for ChatGPT Pro and Business | Monthly API credits | None | Up to $500 per month | Oct 28 |
Anthropic | Claude API (Messages, Token Counting) | Maximum request size | 32 MB | 128 MB | Oct 28 |
Gemini 4 Carbon | Input | New model | $2.00 introductory, then $4.00 | Oct 28 (limited access) | |
Gemini 4 Carbon | Output | New model | $10.00 introductory, then $20.00 | Oct 28 (limited access) | |
Gemini 4 Carbon | Cached input | New model | 95% off the input price | Oct 28 (limited access) | |
Gemini API (Generate Content API) | Request storage, new paid projects | Off by default | On by default (store=true) | Oct 26 | |
Moonshot AI | Kimi K3.1 Pro | Input (cache miss) | New model (Kimi K3: $3.00) | $2.00 | Oct 23 |
Moonshot AI | Kimi K3.1 Pro | Cached input (cache hit) | New model (Kimi K3: $0.30) | $0.20 | Oct 23 |
Moonshot AI | Kimi K3.1 Pro | Output | New model (Kimi K3: $15.00) | $10.00 | Oct 23 |
........
··········
OPENAI PRICES GPT-6.1 TERRA AT HALF THE RATES OF GPT-6.1 SOL.
OpenAI released GPT-6.1 Terra on October 26, describing it in its API changelog as a fast, efficient model for visual reasoning, computer use and natural writing. The official pricing page lists it at $1.00 per million input tokens, $0.05 for cached input and $5.00 for output on the Standard tier. Cache writes are priced separately: $1.25 per million tokens with 30-minute retention and $2.50 with 12-hour retention. On the Batch and Flex tier the rates halve to $0.50 input, $0.025 cached input and $2.50 output. Every one of these figures is exactly 50% of the corresponding GPT-6.1 Sol price, which remains at $2.00 input, $0.10 cached input and $10.00 output.
Two conditions on the model page change the arithmetic for some workloads. Prompts longer than 272,000 input tokens are billed at twice the input and cache rates and 1.5 times the output rate, and that surcharge applies to the whole request, not only to the tokens above the threshold. Regional processing adds a 10% premium where it is available. The context window is 1,050,000 tokens with up to 128,000 output tokens, so long-context use is possible but pays the higher band. Our launch coverage is in OpenAI releases GPT-6.1 Terra with computer use, $1/$5 API pricing and a 1M-token context, and the comparison with competing models is in GPT-6.1 Terra vs Claude Sonnet 5.5, Gemini 4 Argon and Grok 4.7.
The model is also available through cloud resellers. Amazon announced on October 26 that GPT-6.1 Terra is generally available on Amazon Bedrock, stating that its input and output token prices are at least 50% lower than those of GPT-6.1 Sol and GPT-5.6 Terra, but the announcement gives no exact figures and refers to the Bedrock pricing documentation. Microsoft announced general availability in Microsoft Foundry without publishing per-token prices in its post. Buyers on either cloud should read the reseller's own price sheet rather than assume OpenAI's direct rates apply.
··········
MOONSHOT AI PRICES KIMI K3.1 PRO A THIRD BELOW KIMI K3.
Moonshot AI launched Kimi K3.1 Pro on the Kimi API on October 23, alongside a faster variant. Its official pricing page lists the standard model at $2.00 per million input tokens on a cache miss, $0.20 on a cache hit and $10.00 for output. Cache writes are billed separately by retention time: $2.00 per million tokens for a 5-minute TTL and $4.00 for a 1-hour TTL. The kimi-k3.1-pro-highspeed variant costs twice as much for uncached input and output, at $4.00 and $20.00, with the same $0.20 cache-hit rate.
The same page lists Kimi K3 at $3.00 input, $0.30 cached input and $15.00 output, noting that K3 is one-third higher than standard K3.1 Pro on each rate. Pricing is flat, with no tiering by context length across the 1,048,576-token window, and the API documentation states that kimi-k3.1-pro and kimi-k3 batch jobs are billed at 60% of the standard price, which works out to $1.20 input and $6.00 output for K3.1 Pro. The model itself is covered in Moonshot AI launches Kimi K3.1 Pro.
··········
GOOGLE LISTS INTRODUCTORY $2/$10 PRICING FOR GEMINI 4 CARBON.
Google announced Gemini 4 Carbon on October 28 as a model for cybersecurity and long-horizon agent work, and its announcement includes API prices: an introductory $2 per million input tokens and $10 per million output tokens, rising to $4 input and $20 output after the introductory period, with cached input discounted by 95% from the input price. At the introductory rate that discount implies $0.10 per million cached tokens. The announcement does not give an end date for the introductory period, and Gemini 4 Carbon did not yet appear on the Gemini Developer API pricing page when it was checked on October 30, so these figures come from Google's announcement rather than its price list.
Access is the bigger constraint. Google is making Carbon available first to a subset of cyber defenders and plans to add more trusted partners through its Fairwind Program, with a broader release to developers and businesses described only as coming soon. The price structure mirrors Gemini 4 Argon, which the pricing page lists at $2.00 input and $10.00 output through November 30, 2026 and $4.00 and $20.00 from December 1. Details of the launch are in Google launches Gemini 4 Carbon for cybersecurity and long-horizon agents.
··········
TERMS AND LIMITS: GEMINI REQUEST STORAGE, CLAUDE'S 128 MB REQUESTS AND OPENAI'S API CREDITS.
On October 26 the Gemini API release notes recorded a data-logging change: Generate Content API storage is now on by default for new projects on the paid tier, with requests stored as store=true to match the Interactions API. Existing projects keep their current setting. Developers who do not want requests stored can set store=false per request or turn off Generate Content API logging in the Settings panel of the Logs and Datasets dashboard. The change has no price attached, but it matters for teams with data-retention obligations who create new projects. It is explained in Google turns on Gemini API request storage by default for new paid projects.
On October 28 Anthropic raised the maximum request size of the Messages API and the Token Counting API from 32 MB to 128 MB on the first-party Claude API. The limit concerns payload size, not tokens or price, but it allows larger documents and images to be sent in a single request instead of being split. The next day Anthropic launched a Usage and Cost Analytics API in beta and moved the existing Usage and Cost API to legacy status while keeping it supported. Both changes are covered in Claude API raises its request size limit to 128 MB.
Also on October 28, OpenAI's API changelog announced that eligible ChatGPT Pro and Business subscribers can claim up to $500 per month in free API credits by linking an API organization they own. Token prices do not change, but for small teams already paying for those ChatGPT plans the credits offset part of the API bill described in the workload table below. Eligibility and limits are detailed in OpenAI gives ChatGPT Pro and Business subscribers up to $500 a month in API credits.
··········
SCHEDULED CHANGES ALREADY POSTED ON OFFICIAL PAGES.
Three dated changes appear on official pricing pages and take effect soon, but are not counted in this week's table because the notices do not say when they were first published. Moonshot AI's pricing documentation states that from October 31, 2026, K2 series models will no longer be available to newly registered users, that existing users are not affected, and that all K2 series models are planned for retirement by the end of December 2026. DeepSeek's documentation states that its Search and Fetch tools cost $0.75 per 1,000 successful requests, and that calls made through DeepSeek Harness, Claude Code or Codex, currently free, will be billed per request from 00:00 Beijing time on November 1, 2026.
Google's Gemini API pricing page shows that the current Gemini 4 Argon rates are introductory: $2.00 input, $10.00 output and $0.10 cached input through November 30, 2026, doubling to $4.00, $20.00 and $0.20 from December 1. Batch rates double in the same way, from $1.00 and $5.00 to $2.00 and $10.00. Teams with Argon in production should budget for the higher rates now.
··········
WHAT THREE TYPICAL WORKLOADS NOW COST ACROSS THE MAIN MODELS.
To show how the new price lists compare in practice, we priced three workloads on current models from the main providers, using only prices listed on the official pages on October 30, 2026. The support chatbot handles 10,000 requests a day for 30 days; each request sends 3,000 input tokens, of which 2,000 are a cached system prompt, and returns 400 output tokens. The coding agent runs 1,000 sessions of 50 turns each; every turn sends 60,000 input tokens with a 90% cache-hit share and returns 1,500 output tokens. The document job summarizes 1,000 documents a day for 30 days, each with 150,000 input tokens and 2,000 output tokens, without caching and on the batch tier wherever a batch price is published.
Cache-write charges are excluded because they depend on how often a cache is refreshed; they apply to OpenAI, Anthropic and Moonshot AI, which bill cache writes separately. No request crosses GPT-6.1 Terra's 272,000-token surcharge threshold. Claude Haiku 5.5 is priced at its standard band for the chatbot and agent and at its over-100,000-token band for the document job. Kimi K3.1 Pro batch costs use the documented 60% rate. Mistral Large 4 has no published batch rate, so its document job uses standard prices; DeepSeek V4 Pro uses peak prices for the chatbot and agent and off-peak prices for the document job, since a batch job can be scheduled outside peak hours.
........
Model | Support chatbot (per month) | Coding agent (per 1,000 sessions) | Document summaries (per month) |
GPT-6.1 Terra | $930.00 | $810.00 | $2,400.00 |
GPT-6.1 Sol | $1,860.00 | $1,620.00 | $4,800.00 |
Claude Sonnet 5.5 | $1,860.00 | $1,620.00 | $4,800.00 |
Claude Haiku 5.5 | $96.00 | $94.50 | $1,200.00 |
Gemini 4 Argon (to Nov 30) | $1,860.00 | $1,620.00 | $4,800.00 |
Gemini 4 Argon (from Dec 1) | $3,720.00 | $3,240.00 | $9,600.00 |
Kimi K3.1 Pro | $1,920.00 | $1,890.00 | $5,760.00 |
Mistral Large 4 | $504.00 | $556.50 | $3,276.00 (standard) |
DeepSeek V4 Pro | $897.60 (peak) | $811.80 (peak) | $3,088.80 (off-peak) |
........
The arithmetic for GPT-6.1 Terra shows how the figures are built. Each chatbot request costs 1,000 uncached input tokens at $1.00, 2,000 cached tokens at $0.05 and 400 output tokens at $5.00, or $0.0031, and 300,000 requests a month bring that to $930. Each agent turn costs 6,000 uncached tokens at $1.00, 54,000 cached tokens at $0.05 and 1,500 output tokens at $5.00, or $0.0162, so a 50-turn session costs $0.81. Each batch summary costs 150,000 tokens at $0.50 and 2,000 tokens at $2.50, or $0.08, and 30,000 documents a month cost $2,400.
The comparison favours GPT-6.1 Terra most clearly in output-heavy and cache-heavy work, because its $5.00 output rate and $0.05 cache-hit rate are half those of the $2/$10 class shared by GPT-6.1 Sol, Claude Sonnet 5.5 and introductory Gemini 4 Argon. Kimi K3.1 Pro lands slightly above that class in these scenarios because its cached input costs $0.20 rather than $0.10, which weighs heavily in the agent workload where 90% of input is cached. Among the larger models, Mistral Large 4 is the cheapest for the chatbot and the agent in these scenarios, DeepSeek V4 Pro at peak rates costs about the same as GPT-6.1 Terra and half that off-peak, and Claude Haiku 5.5 costs roughly a tenth of GPT-6.1 Terra for short prompts. The table compares price only; the models differ in capability, context limits and features, and a cheaper model that needs more turns or retries can cost more in practice.
··········
WHAT THE WEEK MEANS FOR API BUYERS.
The established facts are clear on the official pages. GPT-6.1 Terra halves the per-token cost of OpenAI's GPT-6.1 Sol across standard, cached and batch rates, Kimi K3.1 Pro is listed a third below Kimi K3, and Gemini 4 Carbon is announced at the same $2/$10 introductory structure as Gemini 4 Argon. Teams running high-volume chat or cache-heavy agents on GPT-6.1 Sol now have an OpenAI option at half the price, provided GPT-6.1 Terra's quality is sufficient for the task and prompts stay below 272,000 tokens.
Several points remain open. Google has not said when Gemini 4 Carbon's introductory period ends or when access widens beyond selected cyber defenders, and the model is not yet on the Gemini API pricing page. Amazon Bedrock and Microsoft Foundry have not published GPT-6.1 Terra's per-token rates in their launch announcements. The scheduled changes at Moonshot AI, DeepSeek and Google, the K2 closure on October 31, DeepSeek tool billing from November 1 and the doubling of Gemini 4 Argon prices on December 1, are the items most likely to change next month's bills.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
