CRHQ has supported Cursor as a model provider for a while, but only for Composer. As of today, six more models are selectable through that same connection, so your agents can run on a much wider range of AI.
There is nothing to install. The new choices simply appear in the model picker.
What is new
Selectable now under provider Cursor, either per session or as an agent's default:
| Model | Maker | Good for |
|---|---|---|
| Gemini 3.1 Pro | Large-context reasoning and analysis | |
| Gemini 3.6 Flash | Fast, inexpensive general work | |
| Gemini 3.5 Flash | Fast, inexpensive general work | |
| Kimi K3 | Moonshot AI | Long-context work and coding |
| GLM 5.2 | Z.ai | Cost-efficient general work and coding |
| Grok 4.5 | xAI, hosted by Cursor | Extended coding tasks |
These join Composer 2.5 and Composer 2.5 Fast, which are unchanged.
To be clear about what we did here: we did not build six new model integrations. Cursor already offers these models, and we surfaced them properly inside CRHQ. If Cursor changes its catalog, the available list can change with it.
How to use them
- Start a new session, or edit an agent's default model.
- Choose provider Cursor.
- Pick any of the new models from the dropdown.
- Optionally set an effort level, from Low to Max. CRHQ maps your choice to whatever thinking level that specific model supports.
- Send your message as normal.
Everything else behaves identically. Delegation, artifacts, skills, and tools all work exactly as they do on any other provider.
One requirement
Your satellite needs a Cursor account connected, under Settings, then AI Models, then Cursor. If it is not connected yet, connect it once and all of these models become available.
Usage is billed through your own Cursor plan, exactly as Composer is today. CRHQ does not resell or mark up model access.
Cost tracking
Per-turn token usage and cost now appear in each session's Costs tab for all of these models. That also closes a gap where Cursor turns previously showed no cost at all.
| Model | Input, per 1M tokens | Output, per 1M tokens |
|---|---|---|
| Gemini 3.1 Pro | $2.00 | $12.00 |
| Gemini 3.6 Flash | $1.50 | $9.00 |
| Gemini 3.5 Flash | $1.50 | $7.50 |
| Kimi K3 | $3.00 | $15.00 |
| GLM 5.2 | $1.40 | $4.40 |
| Grok 4.5 | Included in the Cursor subscription pool | Shows as approximately $0 |
GLM 5.2 is the most cost-efficient of the set. Grok 4.5 draws from your Cursor subscription pool rather than per-token billing, so it reports close to zero, the same way Composer does.
These figures are equivalent token value based on Cursor's published rates. Actual billing happens on your Cursor plan.
Fewer failed turns
Third-party models occasionally return a momentary "at capacity" signal from Cursor. CRHQ now classifies those as overloaded and retries with backoff instead of failing the turn, so a brief provider hiccup is handled transparently rather than ending your agent's work.
What is not included
Two honest exclusions, so you are not left wondering:
- DeepSeek and Qwen are not available. They are not in Cursor's catalog. We can only enable what the provider offers.
- Cursor's "Auto" router is not enabled, deliberately. It picks a model on your behalf, and the model it actually billed cannot be reliably attributed from the CLI output, which would leave a hole in your cost tracking. CRHQ favours explicit model choice.
Availability
| Fact | Value |
|---|---|
| Availability | All CRHQ satellites |
| Live since | 2026-08-12 |
| Client action required | None, beyond connecting a Cursor account if you have not already |
| Billing | Through your own Cursor plan |