GPT-6.1 Sol vs Terra: which to choose

Compare GPT-6.1 Sol and Terra by task, quota cost and reasoning settings, then verify your first API call.

Both models use the Codex channel on AI in IDE. You can keep the same Codex key and base URL; switching models also requires a reasoning effort supported by the model you select.

Which model should I try?

• Complex code changes or tasks with several dependent steps: start with Sol.
• A scoped fix, a rewrite of supplied text, or a clearly defined visual task: try Terra and check the result against your requirements.
• An existing workflow that already works: compare both on the same task, repository context and reasoning effort. Choose based on correctness, time and quota consumed, including retries.

These starting points follow OpenAI's model selection guidance; they are not AI in IDE benchmark results. Terra's provider reference rates are lower, but your quota cost depends on the configured rates below and the tokens your task uses.

Reasoning effort controls how much reasoning the model performs; higher settings can use more time and tokens. Both accept low, medium (API default), high, xhigh and max. Terra also accepts none; Sol does not. The Codex setup guide explicitly uses high.

Compare quota costs

The main table shows quota consumed per 1M tokens, at Standard rates for requests with at most 272K input tokens. It uses AI in IDE's configured model prices × the Codex channel factor. Provider reference prices are shown separately.

Quota is your usage balance, not the amount you pay at checkout. Use the top-up cost calculator, selecting the model and Codex channel, to estimate the purchase needed for a token volume. Check your plan and final checkout price. Cache writes, tools and other service tiers can add or change charges.

272K is the long-input surcharge threshold, not the context limit. Above it, input and cache rates double and output rates become 1.5× for the whole request.

Start with your Codex key

New to AI in IDE: sign in, then open API keys and create a key with the Codex channel. The key and its funding source need a positive balance; view plans if you need quota. Follow the Codex CLI and VS Code setup guide.

Already configured: edit these top-level settings in ~/.codex/config.toml, keeping the provider and authentication from the guide:

Copy
1
model = "gpt-6.1-terra"
2
model_reasoning_effort = "high"

Both models support high, so this example works with model = "gpt-6.1-sol" too. If your Terra configuration uses none, replace it with a Sol-supported effort before switching to Sol. Direct Responses requests use reasoning.effort; Codex CLI uses model_reasoning_effort.

Verify one complete request

This request supplies all the text the model needs. It does not read your local repository. Replace the placeholder with your Codex key and run it in a Bash-compatible shell:

Copy
1
curl --globoff --fail-with-body 'https://api.aiinide.com/responses' \
2
  -H "Authorization: Bearer YOUR_DASHBOARD_API_KEY" \
3
  -H "Content-Type: application/json" \
4
  -d '{"model":"gpt-6.1-terra","reasoning":{"effort":"low"},"input":"Rewrite this sentence in plain English: The application facilitates the modification of account settings.","max_output_tokens":4096}'

Check for status: "completed" and an output_text item inside output[].content[], for example “The app lets you change your account settings.” Wording may vary. Confirm usage in your usage history. This is a billable API request, using your available quota.

If the response is incomplete, inspect incomplete_details.reason; the output limit includes reasoning tokens, so increase it if it was exhausted. For an authentication error, check the key and Codex channel. To test Sol, change the model ID and keep low. For a repository task, use Codex from the project directory or include the relevant files in your API input.

Advanced billing and tool rules

• 30-minute cache writes cost 1.25× the base input rate; 12-hour writes cost 2.5×. Configure prompt_cache_options.ttl or prompt_cache_breakpoint.ttl when needed.
• Fast costs 2× Standard; Flex costs 0.5×. Billing follows the response's service_tier; tiers without configured prices are rejected.
• Built-in web search adds a per-search fee. Hosted Code Interpreter, hosted shell and image generation are unavailable on this channel. These restrictions concern API-hosted tools, not the local files and shell used by Codex CLI.
• Send required conversation and file content with each direct API request. Stored context references, historical audio IDs and remote file or image references are rejected.

The Codex guide covers channel details; the unified API guide covers other compatible clients.

FAQ

Do I need a new key to switch models?

No. Keep your Codex key, provider and base URL. Set the model ID and use a reasoning effort supported by the selected model; none is supported only by Terra.

Which model will cost less for my task?

Compare the current quota rates above and the actual tokens used, including retries. A lower token rate does not guarantee a lower task cost. The provider reference price is not the quota debit or your checkout price.

Can I switch in the middle of a project?

Yes, with a compatible reasoning setting. Keep the project context available to your client. For direct API calls, send the required history each time; the proxy does not store it for you.

Your next step

• Need a key? Sign in, then create a Codex key.
• Have a key? Configure Codex CLI or VS Code and verify one request before running a larger task.
• Planning a budget? Estimate top-up cost, or see Sol and Terra pricing details.