Cursor Basics
GPT-5.6 in Cursor: Luna, Terra, and Sol Explained
OpenAI's GPT-5.6 generation reaches Cursor as three models: Luna is the smallest and cheapest, Terra is the mid-tier middle, and Sol is the flagship for the hardest coding and reasoning. All three share the same agent tools and draw from Cursor's API usage pool. Pick by how hard the task is, not by default.
On this page
What is the GPT-5.6 family in Cursor?
OpenAI's GPT-5.6 generation is available in Cursor as three named sizes rather than a single model. Cursor calls them the Sol, Terra, and Luna family: they share the same tool-calling and agent support, so any of them can drive an agent loop, but they trade intelligence against price. The three cards below map that trade before the pricing detail.
Smallest and lowest-cost of the family.
Built for high-volume, latency-sensitive, or cost-sensitive agent work.
Fast, but with smaller capacity for long, open-ended reasoning.
Mid-tier: sits between Sol and Luna on intelligence and price.
Solid multi-step coding and tool use at roughly half Sol's per-token price.
Faster and cheaper than Sol for routine agent workflows.
OpenAI's flagship GPT-5.6 model in Cursor.
Highest intelligence in the family for challenging coding and reasoning.
Persists through long, multi-hour agent sessions.
All three sit in the OpenAI provider group, which draws from Cursor's API usage pool. That is distinct from Cursor's first-party pool — Auto, 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., and Grok 4.5 — so GPT-5.6 spend is metered separately from the models Cursor trains itself.
This is covered hands-on in Cursor First Hour — 4 short modules, free to read.
How much do Luna, Terra, and Sol cost?
Cursor prices every GPT-5.6 model per million tokens, split into input and output rates. The table lists the published rates from Cursor's models index so you can see the spread from Luna up to Sol at a glance.
- Model
- Luna
- Input / 1M tokens
- $1
- Output / 1M tokens
- $6
- Where it fits
- High-volume loops, subagents, prototyping
- Model
- Terra
- Input / 1M tokens
- $2.5
- Output / 1M tokens
- $15
- Where it fits
- Routine multi-step coding and tool use
- Model
- Sol
- Input / 1M tokens
- $5
- Output / 1M tokens
- $30
- Where it fits
- Hard reasoning, long-running agent sessions
| Model | Input / 1M tokens | Output / 1M tokens | Where it fits |
|---|---|---|---|
| Luna | $1 | $6 | High-volume loops, subagents, prototyping |
| Terra | $2.5 | $15 | Routine multi-step coding and tool use |
| Sol | $5 | $30 | Hard reasoning, long-running agent sessions |
Published per-million-token rates from cursor.com/docs/models. Terra sits at roughly half Sol's per-token price; Luna is the cheapest of the three.
Model rates change as OpenAI and Cursor update the family. Check the current numbers at cursor.com/docs/models before you size a spend cap around them.
Which GPT-5.6 model should I use?
The useful question is how much open-ended reasoning the task needs, not which model is 'best'. Luna and Terra handle most everyday agent work; Sol earns its higher price on the problems that genuinely need more thinking. The table sorts common tasks by that test.
- Task
- High-volume batch edits or a subagent fan-out
- Reach for
- Luna — lowest cost, fast per call
- Task
- A quick prototype or throwaway script
- Reach for
- Luna — good enough, cheapest
- Task
- Routine multi-file feature work with tool calls
- Reach for
- Terra — mid intelligence at mid price
- Task
- A gnarly cross-layer debug or refactor
- Reach for
- Sol — highest intelligence in the family
- Task
- A long, multi-hour agent session
- Reach for
- Sol — persists through long-running work
- Task
- Cost-sensitive work where latency matters
- Reach for
- Luna — built for exactly this
| Task | Reach for |
|---|---|
| High-volume batch edits or a subagent fan-out | Luna — lowest cost, fast per call |
| A quick prototype or throwaway script | Luna — good enough, cheapest |
| Routine multi-file feature work with tool calls | Terra — mid intelligence at mid price |
| A gnarly cross-layer debug or refactor | Sol — highest intelligence in the family |
| A long, multi-hour agent session | Sol — persists through long-running work |
| Cost-sensitive work where latency matters | Luna — built for exactly this |
Match the model to how hard the reasoning is; paying for Sol on a batch loop just burns budget.
Cursor's own notes on Sol flag two behaviors worth knowing: it can over-use subagents on mid-sized tasks, and it sometimes waits for an explicit 'do it' after agreeing with your feedback. On mid-sized work, Terra often gets you there with less ceremony.
What is the Fast mode tier for each model?
Each GPT-5.6 model has a companion Fast tier for priority processing. It is the same model on a faster queue, not a smarter one, and it bills at a premium. The rows below show the naming and the multiplier.
- gpt-5.6-luna-fast
- Luna at priority processing, 2x the standard rates.
- gpt-5.6-terra-fast
- Terra at priority processing, 2x the standard rates.
- gpt-5.6-sol-fast
- Sol at priority processing, 2x the standard rates.
Fast buys queue priority, not more intelligence — the underlying model is unchanged.
How does Sol handle very long context?
Sol is the family member built to keep working across long sessions, and its pricing has a specific rule once a request carries a large prompt. The callout states the threshold and what it does to the rate.
When a Sol request's input exceeds 272k tokens, input pricing doubles and output pricing rises to 1.5x the standard rate. That is the price of feeding it a very large context, so reserve long-context Sol runs for tasks that genuinely need the whole picture rather than defaulting to it.
How does GPT-5.6 usage get billed in Cursor?
GPT-5.6 spend runs through Cursor's API usage pool, and individual plans come with a monthly allowance against it. The pairs below summarize how the metering works.
- Pricing pool
- The API pool — separate from Cursor's first-party pool (Auto, 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., Grok 4.5).
- Included allowance
- Individual Cursor plans include at least $20 of API usage each month.
- Unit
- All rates are per million tokens, split into input and output.
- Fast tiers
- The -fast variant of any model bills at 2x that model's standard rates.
Because the family shares one pool and one included allowance, moving a task from Sol down to Terra or Luna is the most direct way to stretch that monthly budget without leaving the GPT-5.6 generation.
Frequently asked questions
What is the difference between GPT-5.6 Luna, Terra, and Sol?
They are the three sizes of OpenAI's GPT-5.6 family in Cursor. Luna is the smallest and cheapest, built for high-volume, latency-sensitive work. Terra is mid-tier, sitting between the other two on intelligence and price at roughly half Sol's per-token cost. Sol is the flagship — the highest intelligence in the family for hard coding and reasoning and long-running agent sessions. All three share the same agent tools.
How much do the GPT-5.6 models cost in Cursor?
Per million tokens, Cursor lists Luna at $1 input / $6 output, Terra at $2.5 input / $15 output, and Sol at $5 input / $30 output. Each model also has a Fast tier that bills at 2x the standard rates. Verify the current numbers at cursor.com/docs/models before budgeting.
Which GPT-5.6 model should I pick?
Match the model to how much open-ended reasoning the task needs. Use Luna for high-volume loops, subagents, and cost-sensitive work; Terra for routine multi-step coding and tool use; and Sol for hard cross-layer debugging, refactors, and long multi-hour agent sessions. Defaulting everything to Sol just spends more without a matching accuracy gain on easy tasks.
Do GPT-5.6 models count against my first-party Cursor usage?
No. GPT-5.6 Luna, Terra, and Sol are OpenAI models that draw from Cursor's API usage pool, which is metered separately from the first-party pool (Auto, Composer, and Grok 4.5). Individual plans include at least $20 of API usage each month against that pool.
Does Sol cost more when I send a very large prompt?
Yes. When a Sol request's input exceeds 272k tokens, input pricing doubles and output pricing rises to 1.5x the standard rate. Reserve those long-context runs for tasks that need the full picture rather than using them by default.
Sources & last verified
Cursor ships frequently. Facts verified against primary sources on July 15, 2026.