GPT-6.1 Terra vs GPT-6.1 Sol: which one to pick for coding in Codex

OpenAI released GPT-6.1 Terra on October 26, 2026, at half the price of GPT-6.1 Sol: $1 vs $2 per million input tokens, and $5 vs $10 on output. Both are already in Codex CLI. The practical question is whether you can switch your everyday model without losing code quality.

Terra or Sol: which one should you pick for coding?

For hard code, Sol. For short, routine or batch tasks, Terra. OpenAI itself describes Sol as the model "for complex coding, computer use, and professional work" and Terra as "a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing" (official API model pages, checked on November 3, 2026).

In other words, Terra was not positioned as a replacement for Sol in coding. It is the cheap, fast option, and it pays off when the task does not need long planning.

How do GPT-6.1 Terra and GPT-6.1 Sol compare?

Same context window, half the price, and an availability gap that catches anyone using Codex inside the editor.

	GPT-6.1 Terra	GPT-6.1 Sol
API ID	gpt-6.1-terra	gpt-6.1-sol
Release	Oct 26, 2026	Sep 29, 2026
Input / output (1M tokens)	$1 / $5	$2 / $10
Cached input (1M tokens)	$0.05	$0.10
Context window	1,050,000 tokens	1,050,000 tokens
Max output	128,000 tokens	128,000 tokens
Reasoning effort in the API	none to max	low to max (no none)
Codex IDE extension	not listed in the docs	listed
Official recommendation	everyday writing and visual work	complex coding

Sources: OpenAI API model pages (developers.openai.com), Codex models documentation and the Terra system card (deploymentsafety.openai.com), checked on November 3, 2026. Sol release date according to DataCamp.

On both models, a prompt with more than 272K input tokens costs 2x on input and 1.5x on output, for the whole request. The 1M window is there, but filling it changes the math.

What do the benchmarks say about Terra for coding?

On Terminal-Bench 4.0 at max effort, Terra scored 50.3%, vs 21.5% for GPT-5.6 Terra, according to OpenAI's announcement as reported by DataCamp and Developers Digest. For reference, Claude Opus 5.5 is listed at 66.4% on the same benchmark (Anthropic's number, cited by Developers Digest).

We could not find a Sol score on Terminal-Bench 4.0 in any source, so the two cannot be compared side by side on that test. On Terminal-Bench-Science 0.1, DataCamp reports that Terra beats Sol at lower cost, but without the absolute number. All of these results are reported by the vendor itself.

On speed, the Codex docs rate Terra as faster than Sol, without giving tokens per second. Third-party sources say over 3x faster on the same tasks.

How do you use GPT-6.1 Terra in Codex CLI?

Terra requires Codex CLI 0.159.0 or later. Check your version and update if needed:

codex --version
npm install -g @openai/codex@latest

To start a session with Terra, as shown in the official docs:

codex -m gpt-6.1-terra

Inside an open session, type /model, choose the model and, to see the Max and Ultra levels, select More reasoning….

In non-interactive mode, which is where the lower price matters most (scripts, CI, batch jobs):

codex exec -m gpt-6.1-terra "Write unit tests for src/utils/date.ts"

To make it the default, use the same model key from the official example in ~/.codex/config.toml:

model = "gpt-6.1-terra"

Through the API, Terra accepts reasoning.effort set to none, which Sol does not. OpenAI's guide recommends starting at medium and only going up if the quality gain is worth the extra time and tokens:

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gpt-6.1-terra", "reasoning": {"effort": "medium"}, "input": "Explain what this regex does: ^\\d{5}-\\d{3}$"}'

When is each one the right choice?

Use GPT-6.1 Terra when

• The task is short and well defined: a unit test, a docstring, a rename, a regex fix.
• You run codex exec in batch or in CI and the per-call cost adds up.
• The work is visual, like reading a screenshot or comparing a layout.
• You want a near-instant answer with reasoning.effort: none in the API.

Use GPT-6.1 Sol when

• The change spans several files and needs planning before editing.
• The bug is the kind that takes several hypotheses to find the cause.
• You work through the Codex IDE extension, where Terra is not on the official list.
• The session is long: in Codex, Sol uses a 500K-token window by default, and Terra's is not documented.

Can you use both in the same workflow?

Yes, and it is the setup that makes the most sense. Keep Sol as the default for the interactive session and call Terra with -m gpt-6.1-terra for mechanical tasks and codex exec scripts. Since both have the same context window and output limit, switching does not require changing the prompt.