Claude Code vs Cursor: Fable 5.2, Plans & Pricing

Claude Code vs Cursor shifted twice this fall. On October 12, Cursor moved new customers to The X Subscription, where Claude models are billed separately at API rates. On November 9, Anthropic released Claude Fable 5.2, its most capable generally available model, at $8 per million input tokens and $40 output.

The short verdict: Claude Code gives you Anthropic's models on a flat subscription and holds the top spot on the main independent agent leaderboards. Cursor gives you an editor built around parallel agents, cheap in-house models, and a choice of providers, but Claude usage there now costs extra. Below, we compare models, plans, benchmarks, and the CLI, using vendor documentation and named third-party tests.

TL;DR

• Claude Code still defaults to Claude Opus 5.5 ($4/$20 per million tokens) on every paid plan. Claude Fable 5.2 is one /model switch away from v2.1.321. Max and Ultra can spend up to half their weekly limit on it, and Pro pays with usage credits after a free launch week that ends November 16.

• Fable 5.2 is 20% cheaper than Fable 5.1 but still twice Opus 5.5's per-token price. Anthropic's own advice is to start with Opus 5.5 and move up to Fable for the hardest work.

• Cursor added Fable 5.2 the same day, and it tops Cursor's CursorBench 4.0 at 60.0%. Cursor's in-house Composer 2.5 scores 27.7% there, at $0.68 per task against Fable 5.2's $32.79 at max effort.

• Cursor's new plans are Free, Plus at $8, Premium at $30, Super at $100, and Ultra at $200. None includes Claude usage: from Premium up, Claude bills at API list price from prepaid credit. Claude's plans are Pro at $20, Max at $100, and Ultra at $200, the renamed Max 20x.

• Claude Code with Opus 5.5 ranks first on Artificial Analysis's Coding Agent Index (61.7) and on Terminal-Bench 4.1 (65.8%). Neither board currently lists a Cursor agent.

• OpenAI's proposed November 12 shutoff of GPT models in Cursor is still unconfirmed. Claude Code's VS Code extension runs inside Cursor, so using both is an option.

• If you do not write code, neither tool is built for you.

Table of Contents

• The Quick Verdict

• Side-by-Side Comparison Table

• What Changed: Fable 5.2 and Cursor's New Plans

• Claude Code: Strengths and Limits

• Cursor: Strengths and Limits

• Fable 5.2 or Opus 5.5: Which Claude Model to Run

• Benchmarks: What the Numbers Say

• Pricing Comparison

• Claude Code vs Cursor CLI

• When to Pick Claude Code vs Cursor

• What If You Do Not Code at All?

• Claude Code vs Cursor FAQ

• Pick Your Tool, or Skip the Choice

The Quick Verdict

Claude Code is Anthropic's coding agent. It runs in the terminal, a desktop app, the web at claude.ai/code, mobile, VS Code, Cursor and other VS Code forks, and JetBrains IDEs, all sharing one engine and one CLAUDE.md file. It runs Anthropic models only, and your Claude subscription pays for it.

Cursor is an AI code editor built around agents. Its Agents Window runs many agents in parallel, locally or on cloud machines. This fall it added cloud agents that build and host live apps with a database and auth, and voice chat with agents on Grok Voice. Its own models are Grok 4.7 and Composer 2.5, with Claude, Gemini, GPT, and others available on paid plans.

The short answer: pick Claude Code if you want the strongest agent results and subscription billing for Claude models. Pick Cursor if you live in an editor, want several agents running at once, and want cheap in-house models for routine work. If you want Claude inside Cursor, Claude Code's extension bills against your Claude plan, while Cursor's model picker bills the same model at API list price.

Side-by-Side Comparison Table

What Changed: Fable 5.2 and Cursor's New Plans

Four changes since late September matter more than any feature on the list above.

Anthropic released Fable 5.2 and Mythos 5.2

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026. They are the same underlying model. Mythos 5.2 drops some cybersecurity and biology safeguards and is limited to organizations in Anthropic's verification programs, while Fable 5.2 is generally available. Per Anthropic's launch post, Fable 5.2 costs $8 per million input tokens and $40 output, down from $10 and $50 for Fable 5.1, and generates output up to 30% faster.

In Claude Code, Fable 5.2 arrived in v2.1.321 as the new target of the fable alias. It does not replace the default. Opus 5.5 stays the default on every plan, and Anthropic's own guidance is to "start with Claude Opus 5.5, our strongest daily driver," then move up to Fable for harder work. Anthropic's help center says Max, Ultra, and premium Team seats can spend up to 50% of their weekly limits on Fable models, while Pro and standard seats pay with usage credits.

For the launch, Pro and standard Team seats get Fable included through November 16. Max and Ultra subscribers get one to three extra Fable limit resets, depending on how long they have subscribed.

Cursor moved to The X Subscription

On October 12, 2026, Cursor's plans became part of The X Subscription, a bundle of Grok, Cursor, and X Premium. New customers choose Free, Plus at $8, Premium at $30, Super at $100, or Ultra at $200 a month. The old Hobby, Pro, Pro+, Ultra, and Teams plans keep working for existing subscribers, and Cursor calls switching "optional and permanent."

The catch for Claude users is on Cursor's models and pricing page. Free and Plus cannot use third-party models at all. From Premium up, Claude, Gemini, and GPT requests are billed at API list price from prepaid credit, outside the plan's included usage. Cursor says each new plan includes "at least twice as much coding usage with Cursor models" as the legacy plan it replaces, while dropping the included third-party allowance.

The change follows SpaceX's acquisition of Anysphere, Cursor's parent company, which closed on August 14, 2026 at an implied $60 billion valuation. Cursor's docs now call Grok 4.7, from SpaceX's AI unit, "Cursor's flagship model."

Anthropic renamed its plans and filled out the 5.5 family

Anthropic shipped Claude Sonnet 5.5 on September 28 at $2 and $10 per million tokens, and Claude Haiku 5.5 on October 7 at $0.10 and $0.50 for prompts up to 100,000 tokens. On October 30 it renamed Max 5x to Max and Max 20x to Ultra, at the same $100 and $200 prices. Ultra gives 10x Pro's weekly usage, with no five-hour limit for verified accounts.

OpenAI's proposed Cursor cutoff is November 12

OpenAI moved in late August to end Cursor's access to GPT models after the SpaceX deal, proposing November 12, 2026 as the end date. OpenAI's help center still calls that date proposed and says the official date will be confirmed by both companies. GPT-5.6 Sol, Terra, and Luna remain in Cursor's model list for now, and OpenAI has said Cursor will not get newer models such as GPT-6 Astra.

Anthropic went the other way. Co-founder Tom Brown said in August that Anthropic would keep increasing compute for Claude models in Cursor, and Cursor shipped Fable 5.2 on launch day. We cover OpenAI's own coding agent in our Claude Code vs Codex comparison.

Claude Code: Strengths and Limits

Claude Code's CLI works directly on your local files. Opus 5.5, Fable 5.2, and Sonnet 5.5 all have a 1M-token context window, and Anthropic charges one per-token rate across the whole window for all three.

What Claude Code does better:

• Top agent scores. Claude Code with Opus 5.5 leads Artificial Analysis's Coding Agent Index and the Terminal-Bench 4.1 leaderboard, two of the main independent boards for coding agents.

• The strongest model when you need it. Fable 5.2 scores above Opus 5.5 on Anthropic's Terminal-Bench 4.0, CursorBench 4.0, and FrontierCode 2.0 results, and Max and Ultra include it up to half the weekly limit.

• Claude on a subscription. Pro, Max, and Ultra cover Opus 5.5 inside your plan limits. Cursor's new plans bill the same model at API list price.

• Customization depth. Hooks fire on lifecycle events, skills load folder-based instructions on demand, and since October 1, Claude Mods let plugins change deeper behavior.

• Subagents. Each gets its own context window, tools, and model. Since mid-October, the desktop app and VS Code run subagents in the background, and you can set effort per subagent.

• Checkpointing. A snapshot is taken before every change, and /rewind rolls back both files and conversation.

• Cache economics. Opus 5.5 and Fable 5.2 read cache at $0.20 per million tokens, and Sonnet 5.5 at $0.10.

Where Claude Code falls short:

• No free tier. Claude Code is not on Anthropic's Free plan. It starts at $20 a month, while Cursor has a free plan and an $8 one.

• One provider. If Anthropic has a bad week, you have no fallback inside the tool.

• Opaque limits. Anthropic publishes no numeric usage limits for any plan, and Max and Ultra add a separate weekly limit for Fable, so you cannot forecast capacity before buying.

• Fable costs extra on Pro. Outside the launch week, Pro and standard Team seats pay for Fable with usage credits.

• Medium effort by default. Opus 5.5 and Fable 5.2 both start at medium effort in Claude Code, so a hard problem may need /effort high.

• Tokenizer inflation. Anthropic's pricing docs say Claude 4.7 and later models produce "approximately 30% more tokens for the same text," so per-token prices understate the jump from older models.

For how this compares with a managed app-building agent, see CatDoes vs Claude Code.

Cursor: Strengths and Limits

Cursor's bet is throughput: several agents at once, cheap in-house models for routine work, and a menu of frontier models when you need one.

What Cursor does better:

• Parallel work. The Agents Window runs many agents across git worktrees and cloud machines, and work hands off between local and cloud.

• Cheap in-house models. Composer 2.5 costs $0.50 per million input tokens and $2.50 output, 8x less than Opus 5.5. Paid plans include a weekly pool for Composer and Grok.

• Model choice. Fable 5.2, Opus 5.5, Sonnet 5.5, Haiku 5.5, Gemini 4 Argon, GPT-5.6, and Meta's Muse Wonder are all selectable from Premium up.

• A free plan and an $8 plan. Free includes Agent, Chat, and Tab on Cursor's models, and Plus adds more of the same for $8.

• Editor-native features. Tab autocomplete, Design Mode for annotating UI in the built-in browser, a browser tool so the agent can test its own work, and voice chat with agents.

• Cloud agents that ship apps. Since November 5, a cloud agent can build a live app with a database, auth, and hosting, on paid plans with cloud agents enabled.

Where Cursor falls short:

• Claude costs extra. No new plan includes third-party usage. On Teams and Enterprise, Cursor also adds a $0.25 per million token Cursor Token Rate to eligible models such as Opus 5.5 and Fable 5.2, including requests on your own API key.

• Losing GPT models. OpenAI has proposed ending access on November 12, 2026 and says it will not ship future models to Cursor.

• Composer trails at the frontier. It scores 27.7% on Cursor's own CursorBench 4.0, against 56.0% for Opus 5.5 at high effort.

• Pricing churn. Cursor has changed its pricing at least five times since June 2025, most recently with The X Subscription. The first change drew enough backlash for a public apology and refunds.

• Smaller context on in-house models. Composer 2.5 tops out at 200K tokens and Grok 4.7 at 256K in standard mode, against 1M on Claude models.

If you want the managed alternative to this workflow, see CatDoes vs Cursor.

Fable 5.2 or Opus 5.5: Which Claude Model to Run

Both tools now offer the same two top Claude models, so the model choice matters as much as the tool. The data points to Opus 5.5 as the better default and Fable 5.2 as the upgrade for the hardest problems.

Cursor's CursorBench 4.0 leaderboard runs every model in Cursor's harness and reports average cost per task:

At high effort, Opus 5.5 lands within a point of Fable 5.2 for about 41% of the cost per task. Fable 5.2 at max effort posts the top score on the board, but it costs 2.4 times as much as Opus 5.5 at max for 2.2 more points.

Anthropic's own numbers favor Fable 5.2 more. On FrontierCode 2.0, Anthropic says Fable 5.2 at high effort scored about level with Opus 5.5 at max effort "for roughly half the cost per task." Artificial Analysis's general Intelligence Index ranks Fable 5.2 first at 65, against 59 for Opus 5.5.

Early hands-on tests on simpler work found the two models tied on results while Fable 5.2 cost more. In AtomicAgent's single-task test, both passed all 59 hidden tests, with Opus 5.5 at medium effort costing $0.97 per run against $1.87 for Fable 5.2 at medium and $7.70 at high.

Data handling differs too. Fable 5.2 needs 30-day data retention on Anthropic's platform, and Cursor turns it off by default for teams with Privacy Mode and for Enterprise customers. Opus 5.5 is zero-data-retention compatible in Cursor.

The practical rule: default to Opus 5.5, and switch to Fable 5.2 for long, open-ended problems or bugs Opus has already failed on. Keep effort at medium or high unless the task justifies max. For Sonnet and Haiku in the mix, see our breakdown of which Claude model to use.

Benchmarks: What the Numbers Say

Anthropic, Cursor, and independent testers all use different suites and harnesses, so no clean head-to-head exists. Read each table on its own terms.

Anthropic's published board

From the Fable 5.2 launch post and its system card:

Read that board with four caveats. Scores use each model's highest available effort with safeguards on, but Opus 5.5's Terminal-Bench score was measured at xhigh, and GPT-6 Astra's is OpenAI's own figure at high effort. Anthropic reports standard errors of 1.8 points for Fable 5.2 and 2.6 for Opus 5.5 on Terminal-Bench 4.0, so that 1.2-point lead is inside the noise.

Fable 5.2's FrontierCode 2.0 score is at xhigh effort. At max effort it scored 45.7%, or 44.4% with timed-out attempts counted as zero, which is below Opus 5.5. FrontierCode 2.0 is also a rebuilt suite, so these scores are not comparable with the FrontierCode 1.1 numbers in Anthropic's Opus 5.5 launch. And CursorBench is Cursor's own evaluation: Cursor attributes part of Fable 5.2's score to prompt and harness improvements.

Independent agent leaderboards

Artificial Analysis runs each agent with its own model and reports cost per task. On version 2.0 of its Coding Agent Index, launched October 23:

Cursor has not appeared on the index since September 10, when Artificial Analysis moved to Terminal-Bench 4.0, and Artificial Analysis has not said why. Fable 5.2 has not been run yet either. Older figures you may see, such as Composer 2.5 at 62 for $0.07 a task, come from version 1.0 in May. They are not comparable with today's scores, and $0.07 was the non-default standard variant.

The Terminal-Bench 4.1 leaderboard, published November 9, has the same leader: Claude Code with Opus 5.5 at 65.8%, then Claude Code with Sonnet 5.5 at 63.9% and Codex with GPT-6.1 Sol at 60.9%. No Cursor agent or Fable 5.2 run appears there yet. Vals.ai's own Terminal-Bench 4.0 run, in a different harness, has Fable 5.2 at high effort and Opus 5.5 at max tied at 65.15%, at $14.56 and $12.04 per test.

Cursor's own scoreboard

CursorBench 4.0 is the only current board with Cursor's own models on it. Grok 4.7 at extra-high effort scores 46.3% at $6.01 per task, and Composer 2.5 scores 27.7% at $0.68. Opus 5.5 at high effort scores 56.0% for $3.97, so on this suite it costs less than Grok 4.7 and scores almost 10 points higher.

Composer 2.5 is the budget option: about a sixth of Opus 5.5's cost per task at half the score. Its own launch numbers use Terminal-Bench 2.0 and CursorBench 3.1, older suites that are not comparable with the boards above.

Where SWE-bench Verified went

If you came looking for SWE-bench Verified, neither vendor publishes it for current models. Anthropic's Fable 5.2 and Opus 5.5 launch posts do not mention it, and Cursor reports SWE-Bench Multilingual for Composer 2.5 instead. Vals.ai archived its board on September 1 ("we no longer run this benchmark on new model releases"), and BenchLM excludes it from scoring as "contaminated and saturated." Fable 5.2 has no SWE-bench Verified score on any leaderboard we checked.

Pricing Comparison

Subscription plans

Claude's Max and Ultra are monthly only, and Team seats drop to $20 and $100 billed annually. Cursor's annual prices are $80, $300, $1,000, and $2,000 for Plus through Ultra. The multipliers measure different things: Claude's apply to all usage including Opus 5.5, while Cursor's apply only to Cursor's own models. Full rates are on Anthropic's pricing page and Cursor's models and pricing page.

On the common "Claude Pro vs Cursor Pro" question: Cursor Pro ($20) is now a legacy plan that new customers cannot buy. Its closest replacement is Premium at $30, which includes Cursor's own models and lets you buy Claude usage at API rates. Claude Pro at $20 includes Claude Code with Opus 5.5 inside its usage limits.

On "Cursor Ultra vs Claude Max": both companies now sell a $200 plan called Ultra. Claude Ultra gives 10x Pro's weekly usage, no five-hour limit for verified accounts, and up to half the weekly limit on Fable. Cursor Ultra gives 50x Plus usage on Cursor's own models, and any Claude usage is extra. Claude Max at $100 lines up against Cursor Super at $100.

Token pricing

Anthropic's rates are on the Claude platform pricing page; Haiku 5.5 rises to $0.50 and $2.50 for prompts over 100,000 tokens. Cursor's rates apply once you pass the weekly pool, and Cursor says Fast is Composer 2.5's default in the product. Fast mode costs more on Anthropic's side too: $8 and $40 for Opus 5.5, and $16 and $80 for Fable 5.2, paid from usage credits in Claude Code.

What Claude costs inside Cursor

Running the same Claude model in Cursor is now billed differently from Claude Code. On The X Subscription, Claude requests come out of prepaid credit at Anthropic's list price, on Premium and above. On legacy Pro, Pro+, and Ultra plans, they draw from a separate monthly Other Models pool.

On Teams and Enterprise, Cursor adds a $0.25 per million token Cursor Token Rate to eligible third-party requests, including bring-your-own-key. Models whose cache reads cost under $0.10 per million tokens, such as Haiku 5.5, are exempt. If you already pay for Claude Pro or Max, Claude Code's extension in Cursor uses that plan's limits, while Cursor's model picker charges again at API rates.

How each tool meters usage

Claude Code shares one pool with claude.ai: a five-hour window plus a weekly cap, with a separate weekly limit for Fable on Max and Ultra. Verified Ultra accounts have no five-hour limit. Anthropic publishes no numeric limits, and paid plans can turn on usage credits to keep working past them.

Cursor's new plans give one weekly pool for Cursor models, shared with Grok. Past it, you pay from prepaid top-ups, which expire 365 days after purchase. Legacy plans keep two monthly pools, Cursor Models and Other Models, with overage billed in arrears.

Claude Code vs Cursor CLI

Cursor launched its terminal agent in August 2025, and the command is now agent. It has three modes: Agent with full tools, Plan for designing before executing, and Ask for read-only questions. It runs headless with -p for scripts and CI, and you can hand a task to a cloud agent by starting a message with an ampersand.

Claude Code's CLI is the original product. Since September 28, interactive sessions start in auto mode when no permission mode is set. It adds OS-level sandboxing via macOS Seatbelt and Linux bubblewrap, nested subagents with their own context windows, hooks on lifecycle events, and checkpoints that rewind files and conversation together.

The real difference is billing. Cursor's CLI draws from the same weekly pool and prepaid credit as the editor. Claude Code draws from your Claude plan, shared with claude.ai chat, so a heavy afternoon in the terminal eats into your chat capacity, and Anthropic does not publish the limits as numbers.

When to Pick Claude Code vs Cursor

Pick Claude Code when:

• The change is high-stakes, such as auth, payments, or migrations, and a wrong answer is expensive to catch.

• You want Claude models, Fable 5.2 included, covered by a subscription instead of API billing.

• You need a 1M-token context window at one flat rate.

• You want programmable hooks, skills, plugins, and nested subagents for custom governance.

• You already pay for Claude Pro, Max, or Ultra.

Pick Cursor when:

• You want to run several agents in parallel and compare results.

• Most of your work suits cheap in-house models, and you will pay for Claude only on hard tasks.

• You want to start free or at $8, or you want Tab autocomplete and Design Mode alongside agents.

• You want Gemini, Grok, or Meta models next to Claude.

• You want a cloud agent to build and host a live app.

Or run both. Claude Code's VS Code extension installs into Cursor, and your Claude plan covers it. Install it from Anthropic's documentation: on October 29, Cursor warned that malicious copies of the Claude Code extension had appeared in its extension marketplace. A common split is Cursor's models for volume work and Claude Code for the changes that need to be right the first time.

What If You Do Not Code at All?

Claude Code and Cursor both assume you can read code, review diffs, run build commands, and debug errors. They make developers faster. They do not remove the need to be one.

If your goal is a finished mobile app and you have no coding background, an AI app builder is the better fit. It takes a plain-language description and handles the backend, the live preview, and the path to a store-ready build, without a terminal, a diff, or a model picker.

Claude Code vs Cursor FAQ

Which model does Claude Code use by default?

Claude Opus 5.5, on Pro, Max, Ultra, Team, and Enterprise, at medium effort. Fable 5.2 is never the default. You can switch to it with /model on Claude Code v2.1.321 or later, so run claude update if you are behind.

Should you use Fable 5.2 or Opus 5.5 for coding?

Start with Opus 5.5. On Cursor's CursorBench 4.0, Opus 5.5 at high effort scores 56.0% at $3.97 per task, against 56.9% at $9.70 for Fable 5.2 at high effort. Fable 5.2 earns its price on the hardest problems: at max effort it tops that board at 60.0%, and Anthropic recommends moving up to it when Opus 5.5 is not enough.

Can you use Claude Code inside Cursor?

Yes. Anthropic's VS Code extension installs into Cursor and other VS Code forks, and your Pro, Max, or Ultra plan covers it. You can also run claude in Cursor's integrated terminal. Install the extension from the link in Anthropic's documentation rather than searching the marketplace, because Cursor warned on October 29, 2026 that malicious copies had appeared there.

Is Claude Code or Cursor free?

Cursor has a free plan with Agent, Chat, and Tab on Cursor's own models, but no Claude models. Claude Code is not included on Anthropic's Free plan. It starts with Claude Pro at $20 a month, or $17 a month billed annually.

Which tool is better for beginners?

Cursor, because it is free to try and works like a normal code editor, with suggestions appearing as you type. Claude Code assumes comfort with a terminal, though its desktop app and VS Code extension soften that. If you have never written code, neither is the right starting point.

Can Cursor use Fable 5.2 and other Claude models?

Yes. Cursor added Fable 5.2 on launch day and also lists Opus 5.5, Sonnet 5.5, and Haiku 5.5. On The X Subscription, Claude models need Premium ($30 a month) or higher and are billed at Anthropic's list price from prepaid credit. Teams and Enterprise add a $0.25 per million token Cursor Token Rate on eligible models.

What happens to GPT models in Cursor?

OpenAI has proposed ending Cursor's access on November 12, 2026, and says the official date will be confirmed by both companies. GPT-5.6 Sol, Terra, and Luna still appear in Cursor, but Cursor will not get newer OpenAI models. A custom OpenAI API key works only with chat models, so Tab completion stays on Cursor's own models. Cursor says OpenAI models serve about 5% of its traffic.

Which is cheaper, Claude Code or Cursor?

Cursor, if you stick to its own models: Plus costs $8 a month, and Composer 2.5 costs $0.50 per million input tokens against $4 for Opus 5.5. For Claude models the order flips. Cursor's new plans bill Claude at API list price on top of the subscription, while Claude Pro, Max, and Ultra include Opus 5.5 within their usage limits.

Do I need both tools?

Most developers pick one. Since Claude Code runs inside Cursor, using both costs only the two subscriptions, and splitting work by stakes is a reasonable plan: Cursor's models for volume, Claude for the code you cannot afford to get wrong.

Pick Your Tool, or Skip the Choice

Claude Code vs Cursor used to be a question about interfaces. Now it is a question about models and billing. Claude Code gives you Opus 5.5 by default, Fable 5.2 when you need more, and the top spot on the independent agent boards, all on a subscription. Cursor gives you parallel agents, cheap in-house models, and the widest model menu, but on its new plans every Claude request is billed on top.

Both still assume you are a developer. If you are building a mobile app or website and would rather not track which frontier model shipped this week, try CatDoes free. Describe what you want, use the Low or Medium agent tier for routine edits, and switch to High or Max for the hard parts.