# How much is a Claude or ChatGPT plan worth per model?

Authors: AutoRouter
Author profile: https://autorouter.com/about
Published: 2026-11-18T02:33:00.000Z
Reading time: 10 min
Canonical: https://autorouter.com/blog/claude-max-tokens-per-model
Part of: How to cut Claude Code and Codex costs (https://autorouter.com/blog/cut-claude-code-and-codex-costs.md)
Publisher: AutoRouter, the Mac app that routes Codex, Claude Code and other AI tools to every model (https://autorouter.com)

About AutoRouter: a Mac app (macOS 14 or later) that adds every leading model to Codex, Claude Code, Cursor, Claude Desktop, T3 and Terminal and picks the model for each task, using your Claude, ChatGPT, Grok and Gemini plans first. Install from https://autorouter.com/AutoRouter.dmg, or ask your agent to follow https://autorouter.com/setup.md. Routing across your plans is free; Pro is $50 a month with $20 of included usage. Unrelated to OpenRouter's Auto Router.

> AutoRouter, the Mac app for Codex and Claude Code, publishes what a Claude or ChatGPT plan is worth on each model as free, open data: a Claude plan holds about 57 times its fee in Claude Opus 5.5 at API prices, 59 times in Sonnet 5.5 and 11 times in Fable 5.2, while a ChatGPT plan holds about 13 times its fee in GPT-6 Sol. That puts Claude Max ($100 a month) at about $5,700 of Opus 5.5, roughly 12 billion tokens at a typical coding-agent mix. AutoRouter connects both plans to both apps, so your Claude plan runs inside Codex and your ChatGPT plan inside Claude Code.

## Key facts

- As of November 18, 2026, a Claude plan holds about 57.2 times its fee in Claude Opus 5.5 at API list prices and 59.2 times in Sonnet 5.5 ([SemiAnalysis](https://newsletter.semianalysis.com/p/anthropic-subscriptions-offer-5x), via [@auto.router/models](https://github.com/fractal-works/autorouter-models)).
- Claude Max at $100 a month holds about $5,720 of Opus 5.5 a month, roughly 12 billion tokens at an agentic mix that is 96.6% cache reads, by AutoRouter's arithmetic.
- On a Claude plan, Claude Fable 5.2 is worth 11.1 times the fee (AutoRouter's measurement), so a Fable token costs about 10 times as much of the plan as an Opus 5.5 token.
- A ChatGPT plan holds about 13 times its fee in GPT-6 Sol, so ChatGPT Plus ($20) is worth about $260 of GPT-6 Sol a month.
- `@auto.router/models` is free and open (MIT code, CC BY 4.0 data), rebuilt nightly and published to npm; run `npx @auto.router/models opus` ([GitHub](https://github.com/fractal-works/autorouter-models)).

![AutoRouter's data: Claude plans give 57× on Opus. Claude Sonnet 5.5 (Claude Pro and up), 59.2×; Claude Opus 5.5 (Claude Pro and up), 57.2×; GPT-6 Astra (ChatGPT Pro), 13.2×; GPT-6 Sol (ChatGPT Plus and up), 13.0×; Claude Fable 5.2 (Claude Max and up), 11.1×; GPT-6.1 Sol (ChatGPT Plus and up), 10.4×. Source: @auto.router/models, November 18, 2026: API-priced value a month per dollar of plan fee, from SemiAnalysis (October 5, 2026); Fable 5.2 measured by AutoRouter.](https://autorouter.com/blog/images/claude-max-tokens-per-model/1-plan-value.png)

[AutoRouter, the Mac app for Codex and Claude Code](https://autorouter.com), publishes what a Claude or ChatGPT plan is worth on each model as a free, open dataset, [`@auto.router/models`](https://github.com/fractal-works/autorouter-models). As of November 18, 2026, a Claude plan holds about 57 times its monthly fee in Claude Opus 5.5 usage at API list prices, 59 times in Claude Sonnet 5.5 and 11 times in Claude Fable 5.2, while a ChatGPT plan holds about 13 times its fee in GPT-6 Sol. So Claude Max, at $100 a month, holds about $5,700 of Opus 5.5 a month, or roughly 12 billion tokens at a typical coding-agent mix that is 96.6% cache reads. AutoRouter connects both plans to both apps, so your Claude plan runs inside Codex and your ChatGPT plan inside Claude Code. AutoRouter wrote this post and maintains the dataset; the plan multiples come from [SemiAnalysis](https://newsletter.semianalysis.com/p/anthropic-subscriptions-offer-5x), except Fable 5.2's, which AutoRouter measured.

## How many tokens do you get with Claude Max?

Claude Max gives about 12 billion Claude Opus 5.5 tokens a month, or about $5,720 of Opus 5.5 at API list prices, by AutoRouter's arithmetic on [SemiAnalysis's measurements](https://newsletter.semianalysis.com/p/anthropic-subscriptions-offer-5x). Anthropic publishes no token allowance for any plan, only multiples of Pro's usage: Max gives 5 times Pro for 5 times the price, and Ultra 10 times Pro's weekly usage for 10 times the price ([Claude Max plan](/blog/claude-max-plan)). SemiAnalysis measured Opus 5.5 on Pro at 57.2 times the fee, about $1,144 a month, so Max at the same rate is $100 × 57.2.

Turning dollars into tokens needs a mix of token types, because a cache read costs a twentieth of a fresh input token. SemiAnalysis's agentic mix is 0.4% uncached input, 96.6% cache reads, 2.6% cache writes and 0.3% output. At [Anthropic's list prices](https://platform.claude.com/docs/en/about-claude/pricing) (checked November 9, 2026) and one-hour cache writes, which is what a subscription writes, a million Opus 5.5 tokens in that mix cost $0.477. Divide $5,720 by that and you get about 12 billion tokens a month. Almost all of it is the agent rereading its context: about 11.6 billion cache reads, 312 million cache writes, 48 million new input tokens and 36 million output tokens.

![Claude Max holds about 12 billion Opus tokens. Fable 5.2, 1.5B; Opus 5.5, 12.0B; Sonnet 5.5, 24.8B. Source: AutoRouter's math for Max ($100): @auto.router/models plan value, Claude API prices, SemiAnalysis's agentic mix.](https://autorouter.com/blog/images/claude-max-tokens-per-model/2-max-tokens.png)

| Plan, monthly fee | Opus 5.5                    | Sonnet 5.5                  | Fable 5.2                         |
| ----------------- | --------------------------- | --------------------------- | --------------------------------- |
| Pro, $20          | $1,144, about 2.4B tokens   | $1,184, about 5.0B tokens   | Not included (usage credits only) |
| Max, $100         | $5,720, about 12.0B tokens  | $5,920, about 24.8B tokens  | $1,110, about 1.5B tokens         |
| Ultra, $200       | $11,440, about 24.0B tokens | $11,840, about 49.6B tokens | $2,220, about 2.9B tokens         |

Each cell is the plan's fee times the model's multiple from the dataset, and tokens use the same agentic mix. Max and Ultra assume Anthropic's stated 5 and 10 times Pro. One cross-check: SemiAnalysis's own figure for Opus 5.5 on the $200 plan, about $11,726 a month as [Claude Code pricing](/blog/claude-code-pricing) reports, is within 3% of the Ultra cell. Fable 5.2 can use at most half of Max's weekly limit ([Anthropic](https://support.claude.com/en/articles/11049741-what-is-the-max-plan)), and AutoRouter measured it at 11.1 times the fee, with a plausible range of 10.3 to 13.5, from the rate-limit headers of one Ultra account.

## How much is a ChatGPT plan worth per model?

A ChatGPT plan holds about 13 times its fee in GPT-6 Sol at API list prices, so ChatGPT Plus ($20 a month) is worth about $260 of GPT-6 Sol a month and the $100 Pro plan about $1,300. [SemiAnalysis](https://newsletter.semianalysis.com/p/anthropic-subscriptions-offer-5x) found that OpenAI's $100, $200 and $500 Pro plans give the same usage per dollar for every model, and that Plus is comparable for Sol but worse for Astra. Its October 5, 2026 report is titled "Anthropic Subscriptions Offer 5x+ More Value Than OpenAI", and the dataset agrees: Opus 5.5 on a Claude plan is worth 4.4 times as much per dollar of fee as GPT-6 Sol on a ChatGPT plan.

| Model               | Plan value | Lowest plan measured | Measured by                                      |
| ------------------- | ---------- | -------------------- | ------------------------------------------------ |
| GPT-6 Astra         | 13.2×      | ChatGPT Pro ($100)   | SemiAnalysis                                     |
| GPT-6 Sol           | 13.0×      | ChatGPT Plus ($20)   | SemiAnalysis                                     |
| GPT-6.1 Sol         | 10.4×      | ChatGPT Plus ($20)   | SemiAnalysis                                     |
| Other OpenAI models | 10.4×      | None                 | OpenAI's lowest measured value, as an assumption |

GPT-6 Astra and GPT-6.1 Sol are measured in the dataset's [`subscription-value.json`](https://github.com/fractal-works/autorouter-models/blob/main/data/subscription-value.json) but aren't in this month's list of 29 models, so their plan prices aren't in `models.json`. OpenAI says API token prices shouldn't be used to estimate a plan's included tasks ([Codex pricing](https://learn.chatgpt.com/docs/pricing)), so read these as SemiAnalysis's measurements at API prices, not OpenAI's numbers. For each ChatGPT plan's limits in Codex, see [Codex pricing: all Codex plans, prices and usage limits](/blog/codex-pricing).

## What does each model cost on a plan?

On a plan, a model costs its API list price divided by its plan value, which puts Claude Opus 5.5 output at about $0.35 per million tokens on a Claude plan against $20 at list price. That divided price is the `subscription` field in [`@auto.router/models`](https://github.com/fractal-works/autorouter-models), and it is the price AutoRouter assumes for a model when a Claude or ChatGPT plan covers it. Prices below are USD per million tokens, input and output, as of November 18, 2026.

![Opus 5.5 output: $20 listed, $0.35 on a plan. 1. List price ($20 per million output); 2. Divide by plan value (57.2× on a Claude plan); 3. Price on a plan ($0.35 per million output). Source: @auto.router/models, November 18, 2026; Claude API list price checked November 9, 2026; plan value from SemiAnalysis.](https://autorouter.com/blog/images/claude-max-tokens-per-model/3-plan-price.png)

| Model             | List price    | On a plan        | Plan value     |
| ----------------- | ------------- | ---------------- | -------------- |
| Claude Fable 5.2  | $8 / $40      | $0.721 / $3.60   | 11.1×          |
| Claude Opus 5.5   | $4 / $20      | $0.0699 / $0.350 | 57.2×          |
| Claude Sonnet 5.5 | $2 / $10      | $0.0338 / $0.169 | 59.2×          |
| Claude Haiku 5.5  | $0.10 / $0.50 | $0.0090 / $0.045 | 11.1×, assumed |
| GPT-6.2 Astra     | $8 / $40      | $0.769 / $3.85   | 10.4×, assumed |
| GPT-6 Sol         | $2 / $10      | $0.154 / $0.769  | 13.0×          |
| GPT-6.2 Sol       | $2 / $10      | $0.192 / $0.962  | 10.4×, assumed |
| GPT-6.1 Terra     | $1 / $5       | $0.0961 / $0.481 | 10.4×, assumed |
| GPT-6 Luna        | $0.10 / $0.50 | $0.0096 / $0.048 | 10.4×, assumed |

The table shows why the model matters as much as the plan. Fable 5.2 lists at twice Opus 5.5's price, but on the same Claude plan it costs about 10 times as much per token, because the plan gives Fable about a fifth of the value per dollar. Sonnet 5.5 and GPT-6 Sol have the same list price, yet Sonnet costs less than a quarter as much on a Claude plan as GPT-6 Sol does on a ChatGPT plan. "Assumed" marks a model with no measurement of its own, which the dataset gives its vendor's lowest measured value; models from other makers have no plan price at all. To compare models on the API alone, see [Claude API vs subscription: which is cheaper for coding?](/blog/subscription-vs-api-pricing).

## How do you use both plans in Codex and Claude Code?

AutoRouter connects your Claude and ChatGPT plans to both apps, so the plan value above is available in whichever app you work in: your Claude plan runs inside Codex and your ChatGPT plan inside Claude Code, with no keys or config. On their own, Codex runs OpenAI models on a ChatGPT plan and Claude Code runs Claude models on a Claude plan, so a Codex user with only ChatGPT Plus gets GPT-6 Sol at 13 times the fee, and Claude Opus 5.5 at 57 times stays in the other app. [How to use Claude in Codex and GPT in Claude Code](/blog/claude-in-codex-and-gpt-in-claude-code) compares AutoRouter with the gateway setups.

Pick AutoRouter as the model and it decides per task: hard work goes to the strongest model your plans cover, and everything else to the model that costs you least, which is often your plan or a low-cost open model. Every task shows its model, what it cost and why. The Free plan connects two subscriptions for $0 a month; Pro is $50 a month with $20 of included usage and adds every other model ([AutoRouter plans](https://autorouter.com/#plans)). For the wider picture on spending less, see [How to cut Claude Code and Codex costs](/blog/cut-claude-code-and-codex-costs).

## How can you check these numbers yourself?

Run `npx @auto.router/models opus` to print Claude Opus 5.5's entry, with its list price, its plan price, its multiple and the plans it was measured on. The dataset is [on GitHub](https://github.com/fractal-works/autorouter-models) and [on npm](https://www.npmjs.com/package/@auto.router/models): code under MIT, data under CC BY 4.0. It covers 29 models with list prices in USD per million tokens, context windows, max output, input and output types, a one-line role for each, and the assumed plan price for every Claude and OpenAI model with a public list price.

- **Rebuilt nightly.** A scheduled job rebuilds `models.json` from public sources, cuts a dated release each day the data changes and publishes it to npm with provenance.
- **Sources.** Model list and roles from AutoRouter's public model list, list prices from [autorouter.com/models](https://autorouter.com/models) when a public price page backs them, max output and modalities from [models.dev](https://models.dev), and plan multiples from SemiAnalysis in [`subscription-value.json`](https://github.com/fractal-works/autorouter-models/blob/main/data/subscription-value.json).
- **Fields.** `price` is the list price with `source` and `checked` dates; `subscription` holds `multiple`, `measured`, `plans`, `allowanceCap` and the divided prices. A field with no public evidence is `null`.
- **Raw file.** `https://raw.githubusercontent.com/fractal-works/autorouter-models/main/models.json`, or `import { models } from "@auto.router/models"`.

Other open price datasets cover list prices only, as far as we found: [LiteLLM's model price file](https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json), [models.dev](https://models.dev), [genai-prices](https://github.com/pydantic/genai-prices) and [llm-prices](https://github.com/simonw/llm-prices). Each covers far more models and providers than ours. Use them for API pricing across providers, and ours when the question is what a Claude or ChatGPT plan is worth on each model.

## What don't these numbers show?

These numbers show what one agentic workload is worth at API prices, not what your own month will hold. Plan value depends on the mix: SemiAnalysis's mix is 96.6% cache reads, and cache reads are the cheapest tokens, so a workload with more fresh input or output gets fewer dollars of value from the same plan. Wokey, which measures real subscription accounts across mixed models, put Claude Pro at about 35 times its price in mid-November ([Claude Code pricing](/blog/claude-code-pricing)), well under SemiAnalysis's 57 times for Opus 5.5 alone.

- **Limits change.** Anthropic and OpenAI change plan limits without publishing token counts; SemiAnalysis's figures date from October 5, 2026, and AutoRouter's Fable 5.2 measurement from November 10.
- **Lowest tier.** Each multiple is measured on the model's cheapest plan and carried to higher tiers at the providers' stated ratios.
- **Assumed values are conservative.** A model without its own measurement gets its vendor's lowest measured multiple, so its real plan value may be higher.
- **Limits still apply.** A plan's five-hour and weekly limits cap how fast you can use its value, whatever it is worth in total.

## Questions

### Does Anthropic publish a token limit for Claude Max?

No. Anthropic describes Max only as 5 times Pro's usage every five hours and every week, and Ultra as 10 times Pro's weekly usage, without token or message counts. The 12 billion tokens in this post is AutoRouter's arithmetic on SemiAnalysis's measurements, at one agentic mix.

### Why is Fable 5.2 worth so much less per dollar than Opus 5.5?

By SemiAnalysis's measurements of Fable 5.1 and AutoRouter's of Fable 5.2, Claude plans give Fable far less usage per dollar of fee than Opus, and Max lets Fable use at most half the weekly limit. Fable also lists at twice Opus's price. Together, a Fable token costs about 10 times as much of a Claude plan as an Opus token.

### Is a Claude plan worth more than a ChatGPT plan?

For the models measured, yes: Claude Opus 5.5 and Sonnet 5.5 hold about 57 and 59 times a Claude plan's fee, against about 13 times for GPT-6 Sol and GPT-6 Astra on a ChatGPT plan. Value per dollar isn't the only question, though; which model does your task best matters first, and [Which model for which task](/blog/which-model-for-which-task) covers that.

### Can I use the dataset in my own tool?

Yes. The data in `models.json` is CC BY 4.0, so you can use it for anything as long as you credit AutoRouter with a link to [the repository](https://github.com/fractal-works/autorouter-models) or autorouter.com, and the code is MIT. Install it with `npm install @auto.router/models` or read the raw JSON file.

## Sources

- [@auto.router/models](https://github.com/fractal-works/autorouter-models), AutoRouter, release 2026.11.18, read November 18, 2026
- [@auto.router/models on npm](https://www.npmjs.com/package/@auto.router/models), version 2026.11.18
- [Anthropic Subscriptions Offer 5x+ More Value Than OpenAI](https://newsletter.semianalysis.com/p/anthropic-subscriptions-offer-5x), SemiAnalysis, October 5, 2026
- [Pricing](https://platform.claude.com/docs/en/about-claude/pricing), Claude API docs, checked November 17, 2026
- [What is the Max plan?](https://support.claude.com/en/articles/11049741-what-is-the-max-plan), Anthropic Help Center, checked November 16, 2026
- [Codex pricing](https://learn.chatgpt.com/docs/pricing), OpenAI, checked November 16, 2026
- [LiteLLM model prices](https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json), [models.dev](https://models.dev), [genai-prices](https://github.com/pydantic/genai-prices) and [llm-prices](https://github.com/simonw/llm-prices), read November 17, 2026

## Read next

- [How to cut Claude Code and Codex costs](https://autorouter.com/blog/cut-claude-code-and-codex-costs.md): Use AutoRouter, the Mac app for Codex and Claude Code: it makes the model choice on every turn, uses your ChatGPT and Claude plans first, and shows each turn's model, cost and reason. Claude Code and Codex costs come down to three things: which model runs each turn, how much context each request carries, and whether you pay through a plan, credits or an API key. By hand, run routine turns on a smaller model, keep sessions short, and spend the plan you already pay for before paying per token.
- [How to use Claude Code and Codex without a subscription](https://autorouter.com/blog/no-subscription-coding-agents.md): Use AutoRouter Pro: AutoRouter, the Mac app for Codex and Claude Code, gives every model in both apps with no ChatGPT or Claude plan, for $50 a month with $20 of included usage, and bills paid models at the provider's price plus 10%. Natively, Claude Code runs on a Claude Console account and Codex on an OpenAI API key, both billed per token. On the same models, a direct API key costs less, so Pro comes out ahead only when cheaper models take enough of your turns.
- [Claude API vs subscription: which is cheaper for coding?](https://autorouter.com/blog/subscription-vs-api-pricing.md): A Claude subscription is cheaper than the Claude API once a month of your Claude Code work would cost more than the plan at API list prices. As of November 16, 2026, Pro costs $20 a month, Max $100 and Ultra $200, against Anthropic's reported Claude Code average of about $13 per developer per active day at API prices. The API bills each successful request against prepaid credits you buy in the Claude Console, within the rate limits and monthly spend cap of your usage tier.

Tags: [Costs](https://autorouter.com/blog/tags/costs.md), [Claude Code](https://autorouter.com/blog/tags/claude-code.md), [Codex](https://autorouter.com/blog/tags/codex.md)
