Tutorial

How to Run GPT-6.2 Astra and Sol in Parallel with Codex

GPT-6.2 Astra and GPT-6.2 Sol shipped in Codex on November 17, 2026. How to pick each model in the Codex CLI, run several sessions at once in separate Git worktrees, and do the same in Paradev.

Download Paradev

For macOS on Apple Silicon and Intel. Linux AppImage is experimental.

Last updated·

Paradev was called Superset until October 2026.

To run GPT-6.2 Astra or GPT-6.2 Sol in parallel, start one Codex session per task, each in its own Git worktree and branch, and launch each with codex -m gpt-6.2-astra or codex -m gpt-6.2-sol. In Paradev, create one workspace per task, choose Codex as the agent and GPT-6.2 Astra or GPT-6.2 Sol in the model picker, or pass --agent codex --model gpt-6.2-sol to superset ws create. Review each diff before you merge. Worktrees separate working files; they do not sandbox processes or prevent merge conflicts.

What shipped

On November 17, 2026, OpenAI released GPT-6.2 Astra and GPT-6.2 Sol in ChatGPT Work and Codex, and in the OpenAI API. The Codex changelog says: "Choose Astra for your hardest projects, such as implementing and testing a change across an existing codebase," and "Choose Sol for everyday coding, writing, and multistep workflows."

GPT-6.2 AstraGPT-6.2 Sol
ReleasedNovember 17, 2026November 17, 2026
Best forHard, multi-file changesEveryday coding and multistep work
Codex commandcodex -m gpt-6.2-astracodex -m gpt-6.2-sol
Default context in Codex272,000 tokens500,000 tokens
API price per million tokens$8 input, $40 output$2 input, $10 output
Ultrafast modePro $500Pro $200 and Pro $500

Both models have a context window of 1,050,000 tokens in the API. In Codex you can raise the default context to 1 million tokens with the model_context_window setting, per the Codex models page. Requests over 272,000 input tokens cost more in the API: for Astra, OpenAI doubles the input rate and charges 1.5 times the output rate. OpenAI also says Sol completes more tasks on the same paid plan than GPT-6.1 Sol. These figures come from OpenAI and are not independently checked.

The Codex models page does not name one default model. If you set no model, Codex uses "a recommended model," so set the model yourself when it matters which one runs.

Astra or Sol for parallel work

The split maps well onto how parallel work is usually shaped. Most of the tasks you fan out are bounded: tests, small fixes, docs, a migration with a clear spec. Sol costs a quarter of Astra's API price, so it fits those. Keep Astra for the one or two tasks that need the most reasoning, such as a change that spans the codebase. If you are not sure, run the same task on both and compare the diffs.

Pick a model in the Codex CLI

Update Codex to the latest release, then choose the model when you start a session:

Inside a running session, /model switches models. To make one model the default for every new session, set it in ~/.codex/config.toml:

Run several GPT-6.2 sessions at once

Give each task its own worktree and branch, so two sessions never edit the same working files:

Open a terminal in each directory, install dependencies, and start Codex there with the model that fits the task:

Use narrow prompts with a clear stop condition:

Check each session with git branch --show-current before it starts editing.

What worktrees do not separate

Worktrees share repository history and run as your user. They do not give each task its own ports, database, credentials, or API quota. Two sessions can still change the same external service, and their branches can still conflict when you merge. Rerun your checks on the combined branch.

Run GPT-6.2 in Paradev

Paradev is a desktop app that runs Codex and other CLI agents side by side. Each task runs in its own Git worktree and branch, with a terminal, a Changes pane, and a pull request pane next to it. Paradev does not proxy model calls: Codex runs on your own ChatGPT plan or API key. Paradev is open source under the MIT License.

Paradev lists GPT-6.2 Astra and GPT-6.2 Sol in the Codex model picker from version 2.0.10. Codex must be 0.169.0 or later: older Codex releases reject both models with model_not_found.

From the app:

  1. Create a new workspace for each task.
  2. In the composer, choose Codex as the agent and GPT-6.2 Astra or GPT-6.2 Sol in the model picker.
  3. Write the task prompt and start it. Repeat for the next task.
  4. Review each diff in the Changes pane, then commit, push, and open a pull request from there.

From the superset CLI, start a batch of workspaces and group them with a tag:

--effort sets reasoning effort for the launch; both models accept every level up to ultra. On Paradev 2.0.9 or earlier, the picker does not list GPT-6.2: add model = "gpt-6.2-sol" to ~/.codex/config.toml and choose Default model, so Codex reads the model from its config.

Tagged workspaces collect in a sidebar folder with the tag's name. To compare Astra and Sol on the same task, create two workspaces with the same prompt and different --model values. The iPhone and iPad app lets you check on the sessions and review diffs while they keep running on your machine, a host you added, your own Linux server (Bring Your Own Cloud), or a cloud workspace.

Frequently Asked Questions

Should I use GPT-6.2 Astra or GPT-6.2 Sol?

OpenAI recommends Astra for the hardest projects, such as a change across an existing codebase, and Sol for everyday coding and multistep work. Sol costs a quarter of Astra's API price, so it suits most parallel tasks.

Is GPT-6.2 the default model in Codex?

OpenAI's Codex models page does not name a single default. It says Codex uses a recommended model when you set none. Set model in ~/.codex/config.toml or pass -m to be sure which model runs.

Can I use GPT-6.2 with an API key instead of a ChatGPT plan?

Yes. The API model ids are gpt-6.2-astra and gpt-6.2-sol. Astra costs $8 per million input tokens and $40 per million output tokens; Sol costs $2 and $10.

Can I pick GPT-6.2 in Paradev's model picker?

Yes, from Paradev 2.0.10 with Codex 0.169.0 or later. Choose Codex as the agent, then GPT-6.2 Astra or GPT-6.2 Sol in the model picker. On older versions, set the model in ~/.codex/config.toml and choose Default model.

Do parallel GPT-6.2 sessions share files?

Not if each runs in its own Git worktree. They still share ports, databases, credentials, and your model quota, and their branches can conflict when merged.

Where can I read more about running Codex in parallel?

See How to run GPT-6.1 Terra in parallel, the best IDE for Codex, Paradev vs Codex, and Git worktrees for AI agents.