Claude models and API pricing
Four models are current. Prices are the Claude API, standard tier, per million tokens, and the same rates apply when a subscription falls back to usage credits.
| Name | What it is | Use it for | Input | Output |
|---|---|---|---|---|
| Claude Fable 5.1 | The top model, slowest, thinking always on | Hard reasoning, agents that run for hours, when Opus 5.5 fails your evals | $10.00 | $50.00 |
| Claude Opus 5.5 | The model Anthropic tells you to start with | Agentic coding, long documents, most paid-plan chat | $4.00 | $20.00 |
| Claude Sonnet 5.5 | The fast mid-tier model | Default for API apps, coding, agents | $2.00 | $10.00 |
| Claude Haiku 5.5 | The cheap fast model | Classification, extraction, routing, anything at volume | $0.10a | $0.50a |
- a For prompts up to 100,000 tokens. A request whose prompt is longer pays $0.50 input and $2.50 output on every token in it, cache hits included.
All four have a 1M-token context window, a 128K-token maximum output, and training data through June 2026. Claude Mythos 5.1 sits on the rate card at Fable 5.1’s price, but Anthropic sells it only to organizations in its verification programs, so for everyone else it isn’t a choice. The previous generation is still served at its old prices (Opus 5 and Opus 4.8 at $5 / $25, Sonnet 5 at $2 / $10, Sonnet 4.6 at $3 / $15, Haiku 4.5 at $1 / $5), and none of them is cheaper on the rate card than the model that replaced it.
Two discounts stack on top of the rates. Prompt caching charges 1.25x the input price to store a prefix for five minutes and then a fraction of the input price each time it’s read back, and the Batch API halves both input and output for anything that can wait.
| Model | Cache read | Cache write, 5 min | Batch input | Batch output |
|---|---|---|---|---|
| Fable 5.1 | $0.25 | $12.50 | $5.00 | $25.00 |
| Opus 5.5 | $0.20 | $5.00 | $2.00 | $10.00 |
| Sonnet 5.5 | $0.10 | $2.50 | $1.00 | $5.00 |
| Haiku 5.5 | $0.01a | $0.125a | $0.05a | $0.25a |
- a Up to 100,000 prompt tokens. Above that, $0.05, $0.625, $0.25, and $1.25. A one-hour cache write is 2x the input price instead of 1.25x.
Three more lines from the price list: Opus 5.5 has a fast mode at $8 / $40, pinning inference to the US costs 1.1x on every token, and web search is $10 per thousand searches on top of tokens. The names run least to most expensive once you know the order: Haiku, Sonnet, Opus, Fable. The numbers are generations, not sizes, so Sonnet 5.5 is newer than Opus 5, not bigger. For Claude’s rates next to Gemini’s and OpenAI’s in the same units, the field guides hub has the side-by-side table.
Claude Pro vs Max, and the Team and Enterprise plans
Prices are US, per month, from Anthropic’s pricing page and help center. Every plan has a five-hour session limit, paid plans add a weekly limit on top, and chat, Claude Code, and Cowork all draw from the same pool. There is no message count. A long Opus conversation or a Cowork task burns more than a short Sonnet one.
| Free | Pro | Max 5x | Max 20x | Team | Enterprise | |
|---|---|---|---|---|---|---|
| Price/mo | $0 | $20, or $17 billed annually | $100 | $200 | $20 standard seat, $100 premium, billed annually ($25 and $125 monthly) | $20/seat plus usage at API rates |
| Session usage | Base | At least 5x Free | 5x Pro | 20x Pro | 1.25x Pro on standard, 6.25x Pro on premium | Metered, no fixed limit |
| Billing | None | Monthly or annual | Monthly only | Monthly only | Monthly or annual, 2 to 150 seats | Annual, 20-seat minimum |
| Models | Sonnet, Haiku | Opus, Sonnet, Haiku | Same as Pro | Same as Pro | Same as Pro | Same as Pro |
| Fable 5.1 | No | Usage credits only | Up to 50% of the weekly limit | Up to 50% of the weekly limit | Premium seats 50% of weekly limit, standard seats usage credits only | Yes, billed at API rates |
| Claude Code | No | Yes | Yes | Yes | Yes | Yes |
| Cowork | No | Yes | Yes | Yes | Yes | Yes |
| Claude in Chrome | No | Yes | Yes | Yes | Yes | Yes |
| Projects | Up to 5 | Yes | Yes | Yes | Yes, shared | Yes, shared |
| Monthly API credits | No | No | $100 | $200 | $20 per standard seat, $100 per premium, pooled up to $500 | No |
| Good for | Trying it | Everyday work | Frequent use | All-day use | Teams of 2 to 150 | Admin controls at 20-plus seats |
Context windows are 1M tokens for the four current models on every paid plan. Anthropic’s pages disagree here: the Team plan article still says 200k, while the context window article lists 1M for the 5.5 models and Fable 5.1 and the pricing page says “up to 1M, varies by model.” Read 1M for the current models. Older ones range from 1M down to 200k in the context window article, depending on the model and on whether you’re in chat, Claude Code, or Cowork.
Pro is the cheapest plan with Opus, Claude Code, Cowork, and the Chrome extension, and usage credits mean a heavy week costs API rates instead of a forced upgrade. Move to Max 5x when you hit the Pro weekly limit in Claude Code more weeks than not, or when you want Fable without paying per token for it. Anthropic says Max 20x is for people who work with Claude all day. The $100 and $200 of API credits on Max are real money for building on the API, but they do not pay for Claude Code, so don’t count them toward the subscription. Team’s standard seat is Pro with admin tools, a premium seat is a bit more usage than Max 5x (6.25x Pro instead of 5x), and Enterprise is a seat price plus metered usage with a 20-seat floor.
Claude Code pricing, and the other tools
| Name | What it is | Where it runs | What unlocks it | What it costs |
|---|---|---|---|---|
| Claude Code | Coding agent that reads, edits, runs commands, and opens pull requests | Terminal, VS Code and JetBrains, the desktop app, Slack, GitHub Actions | Pro, Max, Team, Enterprise, or a Console account | Plan usage, or API tokens at the rates above |
| Claude Code cloud sessions | The same agent on an Anthropic-run VM that keeps going after you close the laptop | claude.ai/code, the mobile app’s Code tab, the desktop app, or claude --cloud from a terminal | Pro, Max, and Team, or Enterprise with premium or Chat + Claude Code seats | Plan usage, no charge for the VM |
| Claude in Chrome | Extension that reads the page you’re signed in to and clicks, types, and fills forms | Chrome side panel, or called from Cowork and Claude Code | All paid plans | Plan usage |
| Claude Cowork | The non-coding agent: point it at files and tools, come back to a deck, sheet, or memo | Desktop app, web, and mobile, with a built-in browser on desktop | Pro, Max, and Team, or Enterprise when an admin opts in | Plan usage, faster than chat |
| Claude desktop app | Chat, Cowork, and Claude Code in one window, with access to local files and apps | macOS, Windows, Linux (beta, Ubuntu and Debian) | Any plan for chat. Paid for Code and Cowork | Free |
There is no Claude Code price. With a subscription it’s included and draws from the same five-hour and weekly limits as chat. With a Console account it’s pay-per-token at the rates in the first table, which Anthropic says works out to about $13 per developer per active day and $150 to $250 a month across its enterprise deployments, with 90% of users under $30 a day. Pro and Max subscribers who run out can turn on usage credits and keep going at those same API rates. The desktop app, the IDE extensions, cloud sessions, and the GitHub Action all count as Claude Code usage. Fast mode for Opus 5.5 is a research preview at $8 / $40 per million tokens and only works on API billing or usage credits.
Three things are called credits, and each pays for something different.
| Credit | What it pays for | Where it’s used | Expires | Covers Claude Code |
|---|---|---|---|---|
| Usage credits | Usage past your plan’s limits, at API rates | Chat, Claude Code, and Cowork on Pro, Max, and Team | Usually no. In Japan and some other countries, six months after purchase | Yes |
| Monthly API credits | The $100 or $200 a Max plan grants, pooled up to $500 on Team, for the API, Managed Agents, and the Agent SDK | A Console organization you link to the plan, claimable after seven days on it | Yes, at the end of each billing cycle, with no rollover | No |
| Console credits | API usage, bought outright with a card | The API, the Console playground, and anything else billed to a Console account | One year after purchase | Yes, when Claude Code runs on an API key instead of a plan |
Gotchas before you subscribe or build
Fable on Pro is pay-per-token. The pricing page lists Fable under Pro as “usage credits,” which means every Fable token on Pro bills at API rates on top of the subscription. Max and Team premium seats get Fable up to half their weekly limit. If Fable is the reason you’re subscribing, Pro is the wrong plan.
Max’s API credits are not Claude Code money. The $100 and $200 a month go to a Console organization for the API, Managed Agents, and the Agent SDK. They don’t apply to Claude Code, the apps, or usage credits, they don’t roll over, and you need to link a Console org to claim them.
The weekly limit is shared across everything. Session limits reset every five hours, but paid plans also carry a weekly cap across all models, and everything you do in chat, Claude Code, and Cowork counts against the same number. Cowork and agent runs burn through it faster than chat. Both Claude and ChatGPT periodically give out “usage resets” that you can keep for when you hit a limit (Anthropic’s help pages call them limit resets), and in my experience ChatGPT seems more generous with them.
An API key in your shell overrides your subscription. If ANTHROPIC_API_KEY is set, Claude Code bills your Console account at token rates and ignores the plan you’re paying for. Unset it before you log in with your claude.ai account.
Haiku’s price multiplies past 100K tokens. A prompt over 100,000 tokens pays $0.50 / $2.50 instead of $0.10 / $0.50, on the whole request, cache hits included. Above 100K it’s still a quarter of Sonnet’s price, but five times its own short-prompt rate.
Newer models count more tokens. Anthropic says the tokenizer on Claude 4.7 and later produces roughly 30% more tokens for the same text than Sonnet 4.6 and earlier, so a per-token comparison with an older model undercounts the new one’s bill.
Max is monthly only. The annual discount exists on Pro ($17 instead of $20) and on Team seats. Max 5x and 20x bill $100 and $200 every month with no annual option.
Anthropic’s pages disagree about Team’s context window. The Team plan article says 200k. The context window article and the pricing page say up to 1M. The 1M figure in this guide comes from the context window article, and as of the last check the two pages still disagree.
Old Claude Code can’t see Fable. Fable 5.1 in Claude Code needs version 2.1.255 or later, and Cowork needs the current desktop app. If the model picker doesn’t show it, update before you assume your plan excludes it.
Which Claude model to use, by job
| If you want to… | Use | Why |
|---|---|---|
| Chat day to day | Sonnet 5.5 on Free, Opus 5.5 on any paid plan | Opus is what Anthropic says to start with, and Pro includes it |
| Classify, extract, or route at volume | Haiku 5.5, prompts under 100K tokens | $0.10 / $0.50 is the cheapest rate on the card |
| Build an app on the API | Sonnet 5.5, then Opus 5.5 when evals say so | $2 / $10 and $4 / $20. Caching and batch stack on both |
| Run a long agent, or a task Opus keeps failing | Fable 5.1 | $10 / $50. On a plan, only Max and premium seats include it without per-token billing |
| Code in a terminal or editor | Claude Code on Pro | Included. Move to Max 5x when the weekly limit bites |
| Code from your phone or with the laptop closed | Claude Code cloud sessions | Same plan usage, no charge for the VM |
| Hand off reports, spreadsheets, and decks | Cowork on any paid plan | Uses limits faster than chat, so budget for it |
| Click through sites you’re already signed in to | Claude in Chrome | Every paid plan, alone or inside Cowork |
| Process a backlog overnight | Batch API | Half price on input and output |
| Equip a team of 2 to 150 | Team standard seats, premium for the heavy users | $20 a seat annually, with pooled API credits up to $500 |
Sources, all checked October 21, 2026: Anthropic’s pricing page, the API pricing page and models overview in the platform docs, the Claude Code docs on costs and cloud sessions, the product pages for Claude Code, Claude in Chrome, Cowork, and the desktop app, and the help center articles on the Pro plan, Max plan, Team plan, usage credits, monthly API credits, paying for the API, Fable models on your plan, Claude Code on Pro and Max, and context windows by plan.
Changelog
- October 21, 2026. First published. Four current models at $0.10 / $0.50 (Haiku 5.5), $2 / $10 (Sonnet 5.5), $4 / $20 (Opus 5.5), and $10 / $50 (Fable 5.1), with cache and batch rates. Plans at $0, $20, $100, $200, and $20 to $125 a seat. Claude Code, cloud sessions, Claude in Chrome, Cowork, and the desktop app by plan.