Research
Cursor Model Roadmap Tracker
Cursor's May 18, 2026 Composer 2.5 post says Cursor is training a significantly larger model from scratch with SpaceXAI, using 10x more total compute, and a forum recap of Compile repeated it in June. That is a dated roadmap statement, not a claim about the models you can pick today. For today's models, use the current model guide.
On this page
- What has Cursor said about training a larger model?
- How should a team use the model roadmap?
- What should a team watch out for?
- How does the Cursor harness make models cheaper and better?
- Which model should I use for which task?
- What is a sensible default model for each kind of task?
- What if you don't want to choose a model yourself?
- Which current model pages should I check before choosing?
What has Cursor said about training a larger model?
Cursor's statements about the larger model run from April to June 2026: a model-training partnership with SpaceX in April, the from-scratch model and its compute figure in the May Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. post, and a forum recap of Compile in June. Cursor has not announced when the model will be available. The table lists each statement with its source.
- Release fact
- Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. launch
- What to know
- Cursor said Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. improved on sustained long-running tasks, complex instruction following and collaboration behavior.
- Release fact
- April partnership
- What to know
- Cursor's April 21 post announced a partnership with SpaceX on model training, using SpaceXAI's Colossus infrastructure.
- Release fact
- May compute claim
- What to know
- The May 18 Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. post says: "Together with SpaceXAI, we're training a significantly larger model from scratch, using 10x more total compute."
- Release fact
- June forum recap
- What to know
- A June 17 recap of Compile on Cursor's community forum says "Cursor is reportedly training a massive new model from scratch."
- Release fact
- Current selection
- What to know
- These statements are a dated roadmap record. For today's choices, such as Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition., Grok 4.6A Cursor and SpaceXAI model for complex coding and knowledge work, still in the Cursor model pool and succeeded by Grok 4.7; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Grok 4.5, Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition., Claude, Gemini 3.8 FlashGoogle's current speed-tier model in Cursor at $0.75 per million input tokens and $3.75 per million output; it succeeds Gemini 3.7 Flash. Press Enter for the full definition., Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition. and GPT-5.6, use the current model guide.
| Release fact | What to know |
|---|---|
| Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. launch | Cursor said Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. improved on sustained long-running tasks, complex instruction following and collaboration behavior. |
| April partnership | Cursor's April 21 post announced a partnership with SpaceX on model training, using SpaceXAI's Colossus infrastructure. |
| May compute claim | The May 18 Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. post says: "Together with SpaceXAI, we're training a significantly larger model from scratch, using 10x more total compute." |
| June forum recap | A June 17 recap of Compile on Cursor's community forum says "Cursor is reportedly training a massive new model from scratch." |
| Current selection | These statements are a dated roadmap record. For today's choices, such as Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition., Grok 4.6A Cursor and SpaceXAI model for complex coding and knowledge work, still in the Cursor model pool and succeeded by Grok 4.7; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Grok 4.5, Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition., Claude, Gemini 3.8 FlashGoogle's current speed-tier model in Cursor at $0.75 per million input tokens and $3.75 per million output; it succeeds Gemini 3.7 Flash. Press Enter for the full definition., Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition. and GPT-5.6, use the current model guide. |
As of October 17, 2026, unless a row gives a later date. The sources are listed at the end of this page.
This is covered hands-on in Cursor First Hour — 4 short Units, free to read.
How should a team use the model roadmap?
Cursor's release notes and announcements say what changed. They don't say how a team should act on it. The steps below are our suggested order, and the next section lists what to watch out for along the way.
- 1Open the current model guide before you choose a model today.
- 2Read the larger-model announcement as context for how Cursor plans to stand apart over the long term.
- 3Track whether new model availability changes pricing, latency, review quality or agent reliability.
- 4Recheck the official model docs before you quote pricing or benchmark claims.
What should a team watch out for?
Each item below is either a fact about the release, such as its beta status or plan access, or our advice for using it safely. Plan around each one, and recheck the facts when Cursor ships an update.
- Cursor has not announced general availability for the from-scratch model. Treat it as unavailable until Cursor does.
- Keep Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. pricing separate from any future model's pricing.
- Weigh benchmark claims separately from review outcomes on your own repo.
How does the Cursor harness make models cheaper and better?
Cursor doesn't pipe your prompt straight to a model provider. Between your prompt and the model sits Cursor's hosted harness, proprietary cloud infrastructure that handles context selection, caching, retries, compaction and per-model tuning before and after each call.
Interactive diagram. Tab through its regions; each focused region shows its detail in the panel below.
Each layer the harness adds between your prompt and the model.
In an A/B test, loading MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. tools only when the agent needs them cut total agent tokens by about 47% in runs that called an MCP tool. Cursor says the saving varied with the number of MCP servers installed, and that this kind of context work applies to every model in its harness.
Cursor also tunes the harness for each frontier model it supports instead of using one generic pipe, and its field team says the same model performs better there than in a direct call.
Dynamic context discoveryIntroduced in Cursor 2.4
The older approach put more in the window up front, such as every MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. tool definition and long tool output, whether the task needed it or not. That bloats the context and costs tokens, and confusing or contradictory context can lower the quality of the answer. Dynamic context discovery treats context like a filesystem instead. The agent reads from files, tools and MCP output and pulls into the window only what it judges relevant to the task.
- Less irrelevant context can improve answers as well as cutting cost.
- The ~47% figure comes from runs that called an MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. tool, so don't apply it to every session.
- There is no native model memory here. What feels like memory is the harness finding the right context again on each turn.
A Cursor speaker described the company as a harness labCursor as a model-neutral test bed: frontier labs send their newest models to Cursor pre-release to tune them inside the same agent harness every model shares. Press Enter for the full definition.: model providers send their latest models to Cursor before release, so Cursor can test them and build a strong harness for each one.
The speaker made the same point about learning Cursor. New frontier models ship every two to three months, and because Cursor stays model-neutral, you can use whichever frontier model you want behind the same workflow.
Which model should I use for which task?
Start from the current picker, then match the model to the job. High-reasoning choices such as Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition. and GPT-5.6 fit planning, auditing and hard bugs. Faster choices such as Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. and Gemini 3.8 FlashGoogle's current speed-tier model in Cursor at $0.75 per million input tokens and $3.75 per million output; it succeeds Gemini 3.7 Flash. Press Enter for the full definition. fit implementation and repeated lighter work. The current model guide keeps the live roster, pools and task trade-offs in one place. On Plus, the picker holds only Cursor's own models, so plan with Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition. and build with Composer 2.5.
Interactive widget. Tab through its controls; the result updates in the panel below as you change them.
High-reasoning models plan and audit, and faster models execute. Auto routes by the selected mode when you don't want to choose.
- Group
- High reasoning
- Use it for
- Plan modeA mode that makes no edits: it researches the codebase and produces an editable plan you review before any code changes. Press Enter for the full definition., planning, auditing, finding bugs, gnarly production issues.
- Examples
- Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition., GPT-5.6 Sol, Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition.
- Group
- Fast execution
- Use it for
- Building from a plan, edits across a few files, documentation.
- Examples
- Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition., Gemini 3.8 FlashGoogle's current speed-tier model in Cursor at $0.75 per million input tokens and $3.75 per million output; it succeeds Gemini 3.7 Flash. Press Enter for the full definition.
| Group | Use it for | Examples |
|---|---|---|
| High reasoning | Plan modeA mode that makes no edits: it researches the codebase and produces an editable plan you review before any code changes. Press Enter for the full definition., planning, auditing, finding bugs, gnarly production issues. | Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition., GPT-5.6 Sol, Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition. |
| Fast execution | Building from a plan, edits across a few files, documentation. | Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition., Gemini 3.8 FlashGoogle's current speed-tier model in Cursor at $0.75 per million input tokens and $3.75 per million output; it succeeds Gemini 3.7 Flash. Press Enter for the full definition. |
A brain icon next to a model in the picker marks a thinking (reasoning) model. No brain means a standard, faster model.
ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. is Cursor's in-house model, trained specifically on reading and writing code. The rule of thumb at Cursor and outside it: don't use Composer for planning. Use a frontier model to plan, then Composer to execute.
We have a funny quote from my CTO, which is: ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. won't write poetry, it won't do your taxes, but it will write very good code.
Plan with a smart model, build with a fast one
A common pattern in the workflows Cursor's field engineers show is to have a larger, more capable model write a very detailed plan, then let a leaner, faster model build from it. Once the plan is detailed enough, most frontier models execute it correctly, so speed becomes what matters, and speed is what the fast executors optimize for. They say Cursor plans this way internally too. In plan mode, the agent drafts the plan for you to review, then you pick the execution model and click Build to start the phase that writes the code.
When a task does heavy MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. tool-calling, don't put a high-reasoning model on it. You would burn expensive reasoning tokens while it waits on function feedback.
Use an execution model such as Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. instead. It keeps spend under control, and the loop moves faster.
What is a sensible default model for each kind of task?
There's no single right answer, so try candidates on your own codebases and learn where each one gets stuck. Keep your evaluation set current instead of copying a picker from an older workshop. These rough defaults date from October 17, 2026, and assume a plan that includes third-party models:
- Planning and high reasoning: Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition., GPT-5.6 Sol or Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition..
- Balanced daily work: Claude Sonnet 5.5Anthropic's medium-tier model in Cursor at $2 per million input tokens and $10 per million output, with cache reads at $0.10; it keeps Claude Sonnet 5's input and output rates and succeeds it. Press Enter for the full definition. or GPT-5.6 Terra.
- Routine implementation and edits across a few files: Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition..
- Fast, repeated tasks and documentation: Gemini 3.8 FlashGoogle's current speed-tier model in Cursor at $0.75 per million input tokens and $3.75 per million output; it succeeds Gemini 3.7 Flash. Press Enter for the full definition., or Claude Haiku 5.5Anthropic's small, low-cost model in Cursor at $0.10 per million input tokens and $0.50 per million output for requests up to 100k input tokens, with 5x rates above that. Press Enter for the full definition. for the lightest work.
- The hardest long-running work: Claude Fable 5.1Anthropic's highest-priced Cursor model for long, complex work at $10 per million input tokens and $50 per million output, with $0.25 cache reads; Enterprise, Teams and Privacy Mode accounts need data-retention approval to use it. Press Enter for the full definition., but only when its higher cost and data-retention approval are acceptable. Anthropic stores agent input and output for harm-prevention processes.
- A specific behavior, cost or modality: compare Grok 4.6A Cursor and SpaceXAI model for complex coding and knowledge work, still in the Cursor model pool and succeeded by Grok 4.7; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition., Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition., Gemini 4 ArgonGoogle's newest Gemini model in Cursor at $2 per million input tokens and $10 per million output. Press Enter for the full definition. and GPT-5.6 Luna.
- Opus 5, Sonnet 5, Fable 5 and Gemini 3.7 FlashGoogle's previous speed-tier model in Cursor, hidden by default and succeeded by Gemini 3.8 Flash at the same published rates. Press Enter for the full definition. are previous-generation models that Cursor now hides by default.
Judge every model on three axes: performance, cost and speed. Different tasks want different mixes. A Cursor speaker said most customers planning with frontier models are cost-insensitive and reach for the strongest one, while for execution, speed matters most.
For a gnarly bug or a production-down database error, compare runs of a current Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition., GPT-5.6 Sol or Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition. on the same evidence instead of relying on an older family label.
What if you don't want to choose a model yourself?
Two options take the decision off your hands. The first is Auto, which routes the model per request (not on Plus or legacy Start). Since July 22, 2026, on Teams and Enterprise plans, Auto has been powered by Cursor Router, which runs a classifier on each agent request and routes it by task type and complexity. You choose an optimization mode, Cost, Balance or Intelligence, not a model.
Cost keeps the previous Auto routing logic. Cursor's Router docs say it bills as a Cursor-model request, while Balance and Intelligence bill at the routed model's list price. The help center's Router page says all Auto modes bill at that list price, so check your usage before you budget.
The second is Best-of-N, which runs the same prompt with the same inputs across several models in parallel, so you can see which one fits. You can even run one model twice to see how much its output varies. Repeating a prompt costs more up front, but less than redoing a task the wrong model got wrong. It pays off when a wrong first choice is expensive, such as a hard bug or a large refactor. For a small edit, pick a fast model and move on.
One of the most common questions we get at Cursor is: which model should I use for this? Best-of-N is our solution. You can determine it for yourself.
For a high-value debug, some teams turn this into a multi-agent run. They start several current models on the same problem, for example Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition., Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition., Claude Sonnet 5.5Anthropic's medium-tier model in Cursor at $2 per million input tokens and $10 per million output, with cache reads at $0.10; it keeps Claude Sonnet 5's input and output rates and succeeds it. Press Enter for the full definition. and GPT-5.6 Sol. Each one searches the same logs through an MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. such as Datadog, and the team compares their findings. If fully debugging the issue costs a dollar, the spend is worth it.
Which current model pages should I check before choosing?
Use the current model guide first, then open the model-specific docs for availability, pricing and task fit. The table below is the current roster as of October 17, 2026. It replaces older picker lists on this roadmap page, and the June larger-model announcement above stays historical until Cursor publishes a launch.
- Model page
- Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition.
- Practical read
- Current lead Cursor model in the weekly pool: $2 / $6 per million tokens, a 256k context with a 500k variant, and Fast mode at 2x.
- Practical read
- Previous Grok version, still in the Cursor model pool. Compare it against Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition. for deeper long-running work.
- Model page
- Grok 4.5
- Practical read
- Keep for controlled comparisons or policies that still depend on it.
- Practical read
- Cursor's coding model for routine implementation and fast execution.
- Model page
- Claude Sonnet 5.5Anthropic's medium-tier model in Cursor at $2 per million input tokens and $10 per million output, with cache reads at $0.10; it keeps Claude Sonnet 5's input and output rates and succeeds it. Press Enter for the full definition.
- Practical read
- Balanced Claude choice for daily work at $2 / $10 per million tokens. Succeeds Claude Sonnet 5Anthropic's previous medium-tier model in Cursor, hidden by default and succeeded by Claude Sonnet 5.5. Press Enter for the full definition..
- Model page
- Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition.
- Practical read
- High-reasoning Claude choice for difficult planning and debugging at $4 / $20, replacing Claude Opus 5.
- Model page
- Claude Haiku 5.5Anthropic's small, low-cost model in Cursor at $0.10 per million input tokens and $0.50 per million output for requests up to 100k input tokens, with 5x rates above that. Press Enter for the full definition.
- Practical read
- Low-cost Claude for light repeated work at $0.10 / $0.50 per million tokens, up to 100k input tokens.
- Model page
- Claude Fable 5.1Anthropic's highest-priced Cursor model for long, complex work at $10 per million input tokens and $50 per million output, with $0.25 cache reads; Enterprise, Teams and Privacy Mode accounts need data-retention approval to use it. Press Enter for the full definition.
- Practical read
- Highest-capability long-running work at $10 / $50 when the higher cost and data-retention approval are acceptable. Supersedes Claude Fable 5Anthropic's previous top-priced Cursor model, hidden by default and succeeded by Claude Fable 5.1; it carries the same data-retention approval requirement. Press Enter for the full definition..
- Model page
- Claude Sonnet 5Anthropic's previous medium-tier model in Cursor, hidden by default and succeeded by Claude Sonnet 5.5. Press Enter for the full definition., Claude Opus 5, Claude Fable 5Anthropic's previous top-priced Cursor model, hidden by default and succeeded by Claude Fable 5.1; it carries the same data-retention approval requirement. Press Enter for the full definition.
- Practical read
- Previous generation, hidden by default in Cursor's model list. Keep for controlled comparisons.
- Practical read
- Use when image input, UI work or large-context analysis matters.
- Model page
- Gemini 3.8 FlashGoogle's current speed-tier model in Cursor at $0.75 per million input tokens and $3.75 per million output; it succeeds Gemini 3.7 Flash. Press Enter for the full definition.
- Practical read
- Current speed-tier option for lighter repeated tasks and documentation at $0.75 / $3.75 per million tokens.
- Model page
- Gemini 4 ArgonGoogle's newest Gemini model in Cursor at $2 per million input tokens and $10 per million output. Press Enter for the full definition.
- Practical read
- Google's newest Gemini model in Cursor, at $2 / $10 per million tokens.
- Model page
- Gemini 3.7 FlashGoogle's previous speed-tier model in Cursor, hidden by default and succeeded by Gemini 3.8 Flash at the same published rates. Press Enter for the full definition., Gemini 3.6 Flash
- Practical read
- Previous Flash versions, hidden by default. Use them for controlled comparisons and older policies.
- Model page
- GPT-5.6 Luna, Terra and Sol
- Practical read
- Choose the speed, balance or reasoning tier that fits the work.
| Model page | Practical read |
|---|---|
| Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition. | Current lead Cursor model in the weekly pool: $2 / $6 per million tokens, a 256k context with a 500k variant, and Fast mode at 2x. |
| Grok 4.6A Cursor and SpaceXAI model for complex coding and knowledge work, still in the Cursor model pool and succeeded by Grok 4.7; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition. | Previous Grok version, still in the Cursor model pool. Compare it against Grok 4.7Cursor and SpaceXAI's current lead Cursor model, in the weekly pool at $2 per million input tokens and $6 per million output, with a 256k context window and a 500k variant; above 256k input tokens it bills at 2x the standard rate. Press Enter for the full definition. for deeper long-running work. |
| Grok 4.5 | Keep for controlled comparisons or policies that still depend on it. |
| Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. | Cursor's coding model for routine implementation and fast execution. |
| Claude Sonnet 5.5Anthropic's medium-tier model in Cursor at $2 per million input tokens and $10 per million output, with cache reads at $0.10; it keeps Claude Sonnet 5's input and output rates and succeeds it. Press Enter for the full definition. | Balanced Claude choice for daily work at $2 / $10 per million tokens. Succeeds Claude Sonnet 5Anthropic's previous medium-tier model in Cursor, hidden by default and succeeded by Claude Sonnet 5.5. Press Enter for the full definition.. |
| Claude Opus 5.5Anthropic's top Opus-tier model in Cursor at $4 per million input tokens and $20 per million output, 20% below Claude Opus 5, which it replaces. Press Enter for the full definition. | High-reasoning Claude choice for difficult planning and debugging at $4 / $20, replacing Claude Opus 5. |
| Claude Haiku 5.5Anthropic's small, low-cost model in Cursor at $0.10 per million input tokens and $0.50 per million output for requests up to 100k input tokens, with 5x rates above that. Press Enter for the full definition. | Low-cost Claude for light repeated work at $0.10 / $0.50 per million tokens, up to 100k input tokens. |
| Claude Fable 5.1Anthropic's highest-priced Cursor model for long, complex work at $10 per million input tokens and $50 per million output, with $0.25 cache reads; Enterprise, Teams and Privacy Mode accounts need data-retention approval to use it. Press Enter for the full definition. | Highest-capability long-running work at $10 / $50 when the higher cost and data-retention approval are acceptable. Supersedes Claude Fable 5Anthropic's previous top-priced Cursor model, hidden by default and succeeded by Claude Fable 5.1; it carries the same data-retention approval requirement. Press Enter for the full definition.. |
| Claude Sonnet 5Anthropic's previous medium-tier model in Cursor, hidden by default and succeeded by Claude Sonnet 5.5. Press Enter for the full definition., Claude Opus 5, Claude Fable 5Anthropic's previous top-priced Cursor model, hidden by default and succeeded by Claude Fable 5.1; it carries the same data-retention approval requirement. Press Enter for the full definition. | Previous generation, hidden by default in Cursor's model list. Keep for controlled comparisons. |
| Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition. | Use when image input, UI work or large-context analysis matters. |
| Gemini 3.8 FlashGoogle's current speed-tier model in Cursor at $0.75 per million input tokens and $3.75 per million output; it succeeds Gemini 3.7 Flash. Press Enter for the full definition. | Current speed-tier option for lighter repeated tasks and documentation at $0.75 / $3.75 per million tokens. |
| Gemini 4 ArgonGoogle's newest Gemini model in Cursor at $2 per million input tokens and $10 per million output. Press Enter for the full definition. | Google's newest Gemini model in Cursor, at $2 / $10 per million tokens. |
| Gemini 3.7 FlashGoogle's previous speed-tier model in Cursor, hidden by default and succeeded by Gemini 3.8 Flash at the same published rates. Press Enter for the full definition., Gemini 3.6 Flash | Previous Flash versions, hidden by default. Use them for controlled comparisons and older policies. |
| GPT-5.6 Luna, Terra and Sol | Choose the speed, balance or reasoning tier that fits the work. |
Roster checked October 17, 2026 against Cursor's Models & Pricing page. The model docs are the source of truth for availability and cost, so check them again before you quote a price.
Frequently asked questions
Who is this guide for?
Developers and product teams who track which models Cursor offers and what is coming next.
What should I do next?
Check the release facts above against the Cursor sources at the end of this page, then work through the steps in order.
Is Cursor just a wrapper around other models?
No. Cursor's hosted harness handles context selection, caching, retries and per-model tuning. In Cursor's test, loading MCP tools only when needed cut agent tokens by about 47% in runs that called an MCP tool. It also serves Cursor's own Composer models, and Cursor's speakers say providers send their latest models to Cursor before release, so Cursor can build a harness for each one.
Which model should I use for planning versus building?
Use a current high-reasoning choice such as Claude Opus 5.5, GPT-5.6 Sol or Grok 4.7 to write a detailed plan, then a faster executor such as Composer 2.5 to build from it. For tasks with heavy MCP tool-calling, prefer an execution model, so you don't burn reasoning tokens waiting on function feedback. If you would rather not choose, use Auto, or run Best-of-N to compare models on the same prompt. On Plus, which has neither Auto nor third-party models, plan with Grok 4.7 and build with Composer 2.5.
What is Cursor Router?
Since July 22, 2026, Cursor Router has powered Auto on Teams and Enterprise plans. It runs a classifier on each agent request and routes it by task type and complexity. You pick an optimization mode, Cost, Balance or Intelligence, instead of a model. Cursor's Router docs say Cost bills as a Cursor-model request, while Balance and Intelligence bill at the routed model's list price. The help center's Router page says all Auto modes bill at the routed model's list price.
Sources & last verified
- Cursor blog: Introducing Composer 2.5
- Cursor model doc: Composer 2.5
- Cursor model doc: Claude Fable 5
- Cursor model doc: Claude Opus 5
- Cursor model doc: Claude Sonnet 5
- Cursor model doc: Gemini 3.1 Pro
- Cursor model doc: Gemini 3.6 Flash
- Cursor model doc: Gemini 3.7 Flash
- Cursor model doc: GPT-5.6 Luna
- Cursor model doc: GPT-5.6 Terra
- Cursor model doc: GPT-5.6 Sol
- Cursor model doc: Grok 4.5
- Cursor model doc: Grok 4.6
- Cursor model doc: Grok 4.7
- Cursor model doc: Claude Opus 5.5
- Cursor model doc: Claude Sonnet 5.5
- Cursor model doc: Claude Haiku 5.5
- Cursor model doc: Claude Fable 5.1
- Cursor model doc: Gemini 3.8 Flash
- Cursor model doc: Gemini 4 Argon
- Cursor model doc: Muse Wonder
- Cursor docs: Models and pricing
- Cursor Help: The X Subscription
- Cursor blog: Cursor partners with SpaceX on model training (Apr 21, 2026)
- Cursor Community Forum: Compile 2026 announcements
- Cursor product
- Cursor changelog
- Cursor Docs: Cursor Router
- Cursor Help: Cursor Router
- Cursor blog: Dynamic context discovery (Jan 6)
Cursor ships frequently. Facts verified against primary sources on October 28, 2026.