LLM comparison 2026: models and prices at a glance

We update this 2026 LLM comparison page once a month. At its core is a table of the models from Anthropic, OpenAI, Google and Mistral, as the providers list them on their own model and pricing pages on 6 October 2026. Each row gives the release date, the API price per 1M tokens and the source with its date.

The page does not rank anything as "better" or "worse". It tells you what exists, what it costs and how the providers themselves position their models. What works for your tasks can only be shown by a test with your own data.

As of 6 October 2026

TL;DR

• According to list prices of 6 October 2026, the price range per 1M input plus 1M output tokens runs from USD 0.60 (GPT-6 Luna) to USD 60 (GPT-6 Astra and Claude Fable 5.1), a factor of 100.
• In September, ten models appeared or were announced on the providers' pages, among them Claude Opus 5.5 and Sonnet 5.5, GPT-6 Astra, GPT-6.1 Sol and Gemini 3.8 Flash. Gemini 4 Argon is announced, but released only to selected cyber defenders.
• Before you commit, check the price date, the data location and the end-of-life date: the introductory price of Gemini 3.8 Flash ends on 31 December 2026, and according to Anthropic Claude Sonnet 4.5 is scheduled for shutdown in the Claude API on 30 November 2026.

The table: models and prices in October 2026

All prices come from the providers' pages, retrieved on 6 October 2026, and are shown in the currency stated there. Anthropic, OpenAI and Google quote US dollars; Mistral also shows a dollar sign on its model pages. We do not convert currencies; the prices shown are exclusive of taxes. "Input / Output" means the price per 1M tokens in standard operation, without discounts for batch or caching.

Provider	Model	Released	API price input / output per 1M tokens	Availability and note	Source (as of 6 Oct 2026)
Anthropic	Claude Fable 5.1	1 September 2026	10 / 50 USD (cache read 0.25 USD)	Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, Microsoft Foundry; 1M tokens of context	Models overview, Release notes
Anthropic	Claude Opus 5.5	22 September 2026	4 / 20 USD (cache read 0.20 USD)	same platforms as Fable 5.1; according to Anthropic, Opus 5 costs 5 / 25 USD	Models overview, Release notes
Anthropic	Claude Sonnet 5.5	28 September 2026	2 / 10 USD	same platforms; retirement no earlier than 28 September 2027	Models overview, Release notes
Anthropic	Claude Haiku 4.5	15 October 2025	1 / 5 USD	200,000 tokens of context; retirement no earlier than 15 October 2026	Models overview, Release notes, Deprecations
Anthropic	Claude Mythos 5.1	1 September 2026	according to the release notes the same as Fable 5.1: 10 / 50 USD	restricted access only: as of 6 October 2026 only for selected US organisations through trusted-access programmes	Release notes, Anthropic
OpenAI	GPT-6 Astra	September 2026 (per API changelog)	10 / 50 USD	1.05M tokens of context; Fast mode not available with EU data residency	Model page, Changelog
OpenAI	GPT-6.1 Sol	29 September 2026	2 / 10 USD	supports US and EU data residency	Model page, Changelog
OpenAI	GPT-6 Sol	22 September 2026	2 / 10 USD	EU data residency in Standard, Flex and Batch; OpenAI points to GPT-6.1 Sol as the newer Sol model	Model page, Changelog
OpenAI	GPT-6 Luna	22 September 2026	0.10 / 0.50 USD	EU data residency in Standard, Flex and Batch	Model page, Changelog
OpenAI	GPT-5.6 Sol	9 July 2026	4 / 20 USD (promotional price, according to OpenAI until at least 21 November 2026)	still on the price list	Pricing page, Changelog
OpenAI	GPT-5.6 Terra	9 July 2026	2 / 12 USD	still on the price list	Pricing page, Changelog
OpenAI	GPT-5.6 Luna	9 July 2026	0.20 / 1.20 USD	still on the price list	Pricing page, Changelog
Google	Gemini 3.8 Flash	2 September 2026	0.75 / 3.75 USD until 31 December 2026 (introductory price), from 1 January 2027 1.50 / 7.50 USD	Gemini API, generally available (Paid Tier)	Pricing page, Changelog
Google	Gemini 3.1 Pro Preview	19 February 2026	2.00 / 12.00 USD for prompts up to 200,000 tokens, above that 4.00 / 18.00 USD	Preview status	Pricing page, Changelog
Google	Gemini 4 Argon	announced 30 September 2026	introductory price according to Google 2 / 10 USD once available	released only to selected cyber defenders (Fairwind Program), not generally available	Google
Mistral	Mistral Medium 3.5	28 April 2026 (per model page)	1.5 / 7.5 USD	256,000 tokens of context; open weights under Modified MIT	Model page
Mistral	Mistral Large 3	2 December 2025 (per model page)	0.5 / 1.5 USD	256,000 tokens of context; open weights under Apache 2.0	Model page
Mistral	Mistral Small 4	16 March 2026 (per model page)	0.15 / 0.6 USD	256,000 tokens of context; open weights under Apache 2.0	Model page

Long context at OpenAI: for all seven OpenAI rows, the prices shown apply, according to the respective model page, up to 272,000 input tokens per request. Above that, input is charged at double and output at one and a half times the price, for the whole request. For Gemini 3.1 Pro Preview, the surcharge above 200,000 tokens is stated in the row. For the Anthropic models Fable 5.1, Opus 5.5 and Sonnet 5.5, Anthropic states the full context of 1M tokens at the standard price; for Gemini 3.8 Flash and the Mistral models, the pricing and model pages state no surcharge (Anthropic pricing).

The table does not include models for which we could not confirm the release date or the price on the provider's page, for example Claude Haiku 5.5 (announced according to Anthropic, not yet listed in the models overview of 6 October 2026).

What changed in September

The list follows the calendar; each entry names its source.

• 1 September: Anthropic releases Claude Fable 5.1 and Claude Mythos 5.1. The price stays at 10 / 50 USD, and the cache read prices fall to 0.25 USD. According to Anthropic, Fable 5.1 costs an estimated 25 percent less than Fable 5 on typical workloads (Anthropic, Release notes).
• 2 September: According to the Google changelog, Gemini 3.8 Flash is generally available (Gemini API changelog).
• 3 September: According to OpenAI's API changelog, GPT-6 Astra is released (OpenAI changelog).
• 18 September: Google restricts access to the Gemini 2.5 models to users who have already used them. The models are not deprecated; for new projects Google recommends 3.5 Flash-Lite or 3.8 Flash (Gemini API changelog).
• 22 September: Anthropic releases Claude Opus 5.5 at 4 / 20 USD; Opus 5 costs 5 / 25 USD. On the same day, according to OpenAI, GPT-6 Sol and GPT-6 Luna appear (Anthropic, OpenAI changelog).
• 28 September: Anthropic releases Claude Sonnet 5.5 at 2 / 10 USD. According to the release notes, code and configuration written for Sonnet 5 can break in five places, for example with forced tool use (Release notes).
• 29 September: OpenAI releases GPT-6.1 Sol, according to OpenAI with "near-Astra performance" at one fifth of Astra's standard price (Model page, OpenAI news).
• 30 September: Google announces Gemini 4 Argon, initially for selected cyber defenders. On the same day, Anthropic announces that it will shut down Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) in the Claude API on 30 November 2026, and recommends Sonnet 5.5 (Google, Release notes).

If you have hard-coded a model, you should know the end-of-life dates: according to Anthropic, Claude Haiku 4.5 will be retired no earlier than 15 October 2026, and Claude Sonnet 4.5 is scheduled for shutdown in the Claude API on 30 November 2026 (Deprecations).

How the providers position their models

This section reflects how the providers describe their own models. It is not a ranking, and the descriptions are marketing and documentation text from the respective providers.

Anthropic. The models overview describes Fable 5.1 as a model "for demanding reasoning and long-horizon agentic work", Opus 5.5 "for long-running agentic coding and knowledge work", Sonnet 5.5 as "the best combination of speed and intelligence" and Haiku 4.5 as "the fastest model with near-frontier intelligence". In that overview, Anthropic recommends starting with Opus 5.5 when in doubt and using Fable 5.1 if your own tests with Opus 5.5 produce insufficient results even at higher effort settings (Anthropic models overview). According to Anthropic, Mythos 5.1 is the same model as Fable 5.1 with different safeguards and is available only through trusted-access programmes, currently only for selected US organisations (Anthropic).

OpenAI. According to its model page, GPT-6 Astra is "our most capable model for the most demanding work". According to OpenAI, GPT-6.1 Sol delivers "near-Astra performance at a lower cost", with the advice to compare it with Astra on your own tasks. GPT-6 Sol is "built for complex coding and agentic workflows", GPT-6 Luna is "our most efficient model for focused, high-volume tasks" (Astra, GPT-6.1 Sol, GPT-6 Sol, GPT-6 Luna). In its changelog, OpenAI describes the GPT-5.6 family as follows: Sol for peak capability, Terra for a balance of intelligence and cost, Luna for efficient high-volume tasks (Changelog).

Google. According to Google, Gemini 3.8 Flash is the "most intelligent Flash model", built for long-horizon software engineering, autonomous agents and complex enterprise workflows. Google describes Gemini 3.1 Pro Preview as a third-generation Pro model for multimodal understanding, agentic capabilities and "vibe-coding" (Pricing page). For Gemini 4 Argon, Google cites deep reasoning across long workflows and an output token limit of one million tokens (Google).

Mistral. According to Mistral, Medium 3.5 is a multimodal frontier-class model optimised for agentic and coding use. Large 3 is an open general-purpose model with a mixture-of-experts architecture and 675 billion parameters in total, 41 billion of them active. Small 4 combines instruct, reasoning and coding in one model with 119 billion parameters, 6.5 billion of them active (Medium 3.5, Large 3, Small 4).

What the prices mean for a team

List prices per token tell you nothing yet about your bill. It depends on how many tokens your tasks need, whether caching applies and how often an agent runs loops. The following calculation is pure arithmetic on list prices of 6 October 2026, not an offer and not a forecast. For each model it assumes 1M input tokens plus 1M output tokens, without caching, batch discount or surcharges for long prompts.

Model	Input	Output	Total for 1M + 1M tokens
GPT-6 Astra	10 USD	50 USD	60 USD
Claude Opus 5.5	4 USD	20 USD	24 USD
Gemini 3.8 Flash (introductory price until 31 December 2026)	0.75 USD	3.75 USD	4.50 USD
Gemini 3.8 Flash (from 1 January 2027)	1.50 USD	7.50 USD	9.00 USD

Three things can be read from this, without saying anything about quality. First, under this assumption the factor between Astra and the introductory price of Gemini 3.8 Flash is 13.3 (60 divided by 4.50). Second, according to Google, the Gemini price doubles at the start of the new year; if your budget assumes the introductory price, you need to account for the increase in January. Third, GPT-6.1 Sol and Claude Sonnet 5.5 both cost 2 / 10 USD, so 12 USD for the same assumption, which is half of Claude Opus 5.5.

Whether a cheaper model does your task well enough is decided by a test with your own examples. The providers' positioning in the previous section helps you choose the candidates for that test. How seat prices, usage and rollout add up to a team bill is explained in What does AI cost a team in 2026?. And if you do not yet know where your company stands, the free AI Readiness Check helps, with 12 questions in about 5 minutes.

Availability in the EU and in the enterprise

Only statements that are backed by the providers' pages appear here. They do not replace a review by your data protection and legal departments.

• Anthropic: According to the release notes, Fable 5.1, Opus 5.5 and Sonnet 5.5 are available in the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. The models overview likewise lists IDs for Haiku 4.5 on all five platforms. For Fable 5.1, Anthropic states: until the new Enterprise Frontier Safeguards system is available, eligible customers can use Fable 5.1 with zero data retention (Anthropic, Release notes). With Opus 5.5, Anthropic also raised the five-hour usage limits on the Pro, Max, Team and seat-based Enterprise plans (Anthropic).
• OpenAI: According to OpenAI, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna support EU data residency in Standard, Flex and Batch; GPT-6.1 Sol supports data residency only in the US and EU. Fast mode is not available with EU data residency for Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna, and the Ultrafast mode for Astra introduced on 29 September runs only with global processing and US data residency. For endpoints with regional processing, OpenAI charges a surcharge of 10 percent. According to OpenAI, the data residency feature is not open to every account: the provider refers you to its sales team, and for regions outside the US you need approval for abuse monitoring controls and an addendum on data retention (Data residency guide, Pricing page). The model page of GPT-6 Astra contains no statement on EU residency; the table in the guide lists gpt-6-astra for Responses, Chat Completions and Batch in the US and Europe regions. Consult the table yourself for your use case.
• Google: Gemini 3.8 Flash is generally available in the Gemini API. According to the pricing page, content in the Paid Tier is not used to improve the products, while in the free tier it is. For users in the EEA, Switzerland and the United Kingdom, however, the terms of service state that the data protection terms of the Paid Services also apply to free services and unpaid API quotas (Pricing page, Terms of service). Gemini 4 Argon is not generally available: Google wants to open it to developers, enterprises and consumers "as soon as possible" and names no date (Google).
• Mistral: Mistral offers regional inference through its own EU endpoint (api.eu.mistral.ai) with a surcharge of 10 percent on the list price. According to Mistral, anyone using the global endpoint gets no commitment to a specific processing location. According to Mistral, control-plane data such as account settings, API keys and billing can still be processed outside the selected region (Regional inference). Large 3 and Small 4 are available as open weights under Apache 2.0, Medium 3.5 under Modified MIT (Model pages).

For everyday teamwork, the plan through which a model is offered also matters. You will find the comparison of the enterprise plans of ChatGPT, Copilot and Claude (updated in October 2026) in ChatGPT, Copilot and Claude compared for enterprises. If you are mainly interested in working with a coding agent: What is Claude Code?

How we maintain this page

We update the table monthly; each row carries its source date. The next update is planned for November 2026. Three rules apply:

• A row enters the table only if the release date and the price are stated on the provider's page. If either is missing, we leave the model out.
• Prices are shown in the currency of the provider's page, without conversion. Promotional prices state their confirmed end date or their guaranteed minimum term.
• We do not sort by performance and do not adopt benchmark rankings. We reproduce the providers' descriptions as such.

FAQ

Which AI model is the best?

There is no general answer. It depends on the task, on the price you are willing to bear per request, and on what you require in terms of data location and data processing. The providers themselves position their models by task, and in the Opus 5.5 announcement Anthropic writes that benchmark gaps at this performance level have become a less reliable measure of real differences. So test two or three candidates with your own examples.

What does GPT-6 Astra cost?

According to the OpenAI model page of 6 October 2026, GPT-6 Astra costs USD 10 per 1M input tokens and USD 50 per 1M output tokens (Standard). Cached input costs USD 1 per 1M tokens. For prompts above 272,000 input tokens, double the price applies to input and one and a half times to output for the whole request; Batch and Flex cost half.

What is Claude Fable 5.1?

According to Anthropic, Claude Fable 5.1 is the successor to Fable 5, released on 1 September 2026, for demanding reasoning and long-running agentic work. It costs USD 10 input and USD 50 output per 1M tokens, cache read USD 0.25. According to Anthropic, Claude Mythos 5.1 is the same model with different safeguards and is accessible only through trusted-access programmes, currently only for selected US organisations.

What is the difference between Opus 5.5 and Sonnet 5.5?

According to Anthropic, both have 1M tokens of context and up to 128,000 tokens of output. Opus 5.5 costs 4 / 20 USD per 1M tokens and is described for long-running agentic coding and knowledge work; Sonnet 5.5 costs 2 / 10 USD and is described as the best combination of speed and intelligence. Anthropic recommends starting with Opus 5.5 when in doubt. A test shows which model suits your tasks.

Is Gemini 4 available?

Not generally. Google announced Gemini 4 Argon on 30 September 2026 and is initially opening it only to selected cyber defenders through the Fairwind Program. For the later launch, Google names an introductory price of USD 2 input and USD 10 output per 1M tokens, but no date for broad availability.

Which AI models exist in 2026?

The table above lists 18 entries from four providers that Anthropic, OpenAI, Google and Mistral name with a date and a price in early October 2026: at Anthropic the Fable, Opus, Sonnet and Haiku series, at OpenAI the GPT-6 models Astra, Sol and Luna plus the GPT-5.6 family, at Google Gemini 3.8 Flash and 3.1 Pro Preview, at Mistral Medium 3.5, Large 3 and Small 4. There are further providers and models that we deliberately do not list here because we could not confirm the date and price.

Sources

All sources were retrieved on 6 October 2026.

• Anthropic, Models overview: platform.claude.com/docs/en/models/overview
• Anthropic, Release notes: platform.claude.com/docs/en/release-notes/overview
• Anthropic, Model deprecations: platform.claude.com/docs/en/about-claude/model-deprecations
• Anthropic, Introducing Claude Fable 5.1 and Claude Mythos 5.1 (1 September 2026): anthropic.com/claude-fable-and-mythos-5-1
• Anthropic, Claude Opus 5.5 (22 September 2026): anthropic.com/news/claude-opus-5-5
• OpenAI, API Changelog: developers.openai.com/api/docs/changelog
• OpenAI, API Pricing: developers.openai.com/api/docs/pricing
• OpenAI, Data residency (Your data): developers.openai.com/api/docs/guides/your-data
• OpenAI, model pages: GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna
• OpenAI, news feed (Introducing GPT-6.1 Sol, 29 September 2026): openai.com/news/rss.xml
• Google, Gemini API Pricing: ai.google.dev/gemini-api/docs/pricing
• Google, Gemini API Terms: ai.google.dev/gemini-api/terms
• Anthropic, Pricing: platform.claude.com/docs/en/about-claude/pricing
• Google, Gemini API Changelog: ai.google.dev/gemini-api/docs/changelog
• Google, Gemini 4 Argon (30 September 2026): blog.google/.../gemini-4-argon
• Mistral, model pages: Medium 3.5, Large 3, Small 4
• Mistral, Regional inference: docs.mistral.ai/inference/regional-inference