Cursor’s Best OpenAI Model Costs $7.98 Per Coding Test. The GPT-6.1 Sol It Never Got Costs $1.72 and Scores 55.05% to 37.88%.

The cutoff, in 4 facts

On August 28, two weeks after SpaceX closed its $60 billion purchase of Cursor, OpenAI said it would wind down the contract that supplies its models to Cursor. Here is what is on the record, as reported by CNBC and InfoWorld:

• The date: OpenAI’s proposed shutoff is November 12, 2026. It is a proposal, and the exact day could still move.
• No new models: OpenAI said it won’t give Cursor future models during the wind-down.
• The reason: OpenAI said it “cannot be confident that SpaceX will use our technology within our terms of service.”
• The size: Cursor CEO Michael Truell said on X that OpenAI models serve about 5% of Cursor user traffic, and that Cursor was talking with OpenAI to resolve it.

The second fact is the one that cost Cursor users money. The freeze kept Cursor on the GPT-5.6 family while OpenAI shipped GPT-6 and then GPT-6.1 everywhere else.

What Cursor sells today vs. what OpenAI sells

Cursor’s Models & Pricing page lists 3 OpenAI models: GPT-5.6 Luna, Sol and Terra. No GPT-6 model appears on it. OpenAI’s own pricing page now lists GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna as its flagship tier. Line them up tier by tier:

Cursor prices from cursor.com/docs/models; OpenAI prices from OpenAI’s API pricing page, standard speed. Both opened November 7, 2026. GPT-5.6 Sol’s $4 / $20 is a promotional price that OpenAI says runs until November 30, 2026.

Every newer model costs half or less per token than the one it replaces, on input, cache reads and output. GPT-6.1 Sol’s cache reads cost a quarter of GPT-5.6 Sol’s. For an agent that re-reads the same repository context on every step, cache reads are most of the bill.

The benchmark: 55.05% vs. 37.88%

Vals.ai runs every model through Terminal-Bench 4.0 with the same harness, Mini-SWE-agent, and publishes accuracy, cost, tokens and wall time per test. Its board was last updated November 5, 2026.

*Our division: Vals.ai’s cost per test divided by its accuracy. Everything else is as published on the Vals.ai Terminal-Bench 4.0 page.

GPT-6.1 Sol scores 17.17 points higher than GPT-5.6 Sol and costs 4.6x less per test. Per solved test, the gap is $3.12 vs. $21.07.

The price cut explains only half of that. The other half is token burn: GPT-6.1 Sol used 2.86 million tokens per test, GPT-5.6 Sol used 10.49 million. Half the price per token times roughly a quarter of the tokens is how a 2x price cut turns into a 4.6x cost cut.

Where GPT-5.6 Sol sits against everything else

Cursor users won’t be left without models. The question is which one replaces GPT-5.6 Sol. Here are the other models Cursor lists, on the same Vals.ai board:

*Our division of cost per test by accuracy. Accuracy and cost from Vals.ai, November 5, 2026.

GPT-5.6 Sol is not a bargain inside Cursor either. Claude Sonnet 5.5 scores 26.26 points higher for $2.77 more per test, and costs less per solved test. Claude Haiku 5.5 scores 1.52 points lower at $3.12, less than half the cost.

Grok 4.7, the model family Cursor’s $8 Plus plan is limited to, scores 28.79% at $18.01 per test on this board.

Can you bring your own OpenAI key?

Partly. Cursor’s bring-your-own-key page says OpenAI keys cover “standard, non-reasoning chat models,” and that the model picker shows which ones. Three more rules from the same page:

• Custom keys only work with chat models. Tab keeps using Cursor’s built-in models.
• On individual plans, own-key requests don’t draw from your plan, and OpenAI bills you directly.
• On Teams and Enterprise, eligible third-party requests carry a $0.25 per million token Cursor Token Rate, own key included. Models with a cache-read price strictly below $0.10 per million are exempt.

That exemption has a sharp edge. GPT-6.1 Terra’s cache reads cost $0.05, so it would skip the rate. GPT-6.1 Sol’s cost exactly $0.10, which is not below $0.10, so it would pay it, if your key can reach it in Cursor at all. Check the picker before you plan a team migration around it.

Where GPT-6.1 Sol does run

From OpenAI’s API and Codex pricing pages and GitHub’s supported-models reference, opened November 7, 2026.

What this means if you code with AI

Check the version, not the brand. “OpenAI models in Cursor” and “OpenAI models” stopped meaning the same thing in August. If you picked GPT in Cursor’s menu this fall, you got a model its successor beats by 17.17 points on Terminal-Bench 4.0.

Price per test beats price per token. GPT-6.1 Sol is 2x cheaper per token and 4.6x cheaper per test. Terra is the reverse case: more tokens, still cheaper. List prices tell you half the story.

Move before November 12, not on it. Rules, prompts and agent instructions tuned for GPT behave differently on Claude or Gemini. Re-run your own hardest tasks on the replacement this week, while the old model still answers.

Keep your keys portable. This is the second time in 2 years a coding editor lost a model vendor: Anthropic limited Windsurf’s direct Claude access in June 2025. A setup where you own the API key and can switch tools in a day is cheap insurance.

Full disclosure: we sell an AI coding agent. On MegaBrain, you can bring a key from any provider, or a local model, on every plan, Free included, and it never counts against a limit. Claude, GPT and Gemini come from credits from Pro. See the Cursor alternative comparison or the plans. MegaBrain is not affiliated with Cursor or OpenAI.

Fun facts

GPT-5.6 Sol’s promotional $4 / $20 price runs until November 30, 2026, per OpenAI’s pricing page. In Cursor, the model is set to disappear 18 days before the promotion ends.

GPT-6.1 Sol finished each Terminal-Bench 4.0 test in 45m12s on average, GPT-5.6 Sol in 33m46s. The cheaper, stronger model is the slower one.

On the same board, Kimi K3.1 Pro scores 63.64% at $4.57 per test, 1.51 points behind Claude Opus 5.5 at about 38% of its cost.

Sources

Pages opened November 7, 2026. OpenAI’s announcement blocked automated access, so its dates and quotes come from CNBC’s and InfoWorld’s reporting.

• Vals.ai — Terminal-Bench 4.0 leaderboard (Mini-SWE-agent; accuracy, cost, tokens and duration per test; updated November 5, 2026)
• Cursor Docs — Models & Pricing (per-token prices for GPT-5.6 Luna, Sol and Terra; Cursor Token Rate)
• Cursor Docs — Bring your own API key (supported OpenAI models, Tab, the $0.25 Cursor Token Rate and its exemption)
• Cursor — Pricing (Plus, Premium, Super, Ultra; third-party models at API list price)
• OpenAI API docs — Pricing (GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna, GPT-5.6 Sol and its promotional price)
• OpenAI Codex docs — Pricing (which models each ChatGPT plan includes in Codex)
• GitHub Docs — Supported AI models in GitHub Copilot (GPT-6.1 Sol and Terra, GA)
• CNBC — OpenAI to end model access to Cursor after acquisition by Elon Musk’s SpaceX (August 29, 2026; November 12 date, 5% of traffic)
• InfoWorld — Cursor customers will lose access to OpenAI coding models in November (August 31, 2026)
• OpenAI — Our decision on Cursor following its acquisition by SpaceX (as cited by CNBC and InfoWorld)