Codex CLI 0.171.0 makes GPT-6.2 Sol the catalog default: a config.toml snippet for a third-party base_url you can copy
As of November 20, 2026, rust-v0.171.0 adds GPT-6.2 Sol and Astra to the model catalogs, with Sol as the default catalog selection. To run it on a custom provider, put model = "gpt-6.2-sol" and model_provider = "qcode" at the top of your user-level ~/.codex/config.toml, then set name, base_url, wire_api = "responses" and env_key = "QCODE_API_KEY" under [model_providers.qcode].
Updated 2026-11-20
Point this tool at the QCode endpoint — one key for Claude, GPT and Chinese models.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Four values to memorise first
Sol became the catalog default on release day
rust-v0.171.0 shipped on 2026-11-20 and adds GPT-6.2 Sol and Astra to the model catalogs with Sol as the default catalog selection; with no model key at the top level you get that default, so 0.171.0 lands you on Sol; on a custom provider, treat the model that /model reports inside an interactive session as the one actually in use.
OpenAI official Standard rates (per 1M tokens)
Checked on the official pricing page on 2026-11-20: gpt-6.2-sol bills $2.00 input, $10.00 output, $0.10 cached input, $2.50 for a 30m cache write and $5.00 for a 12h cache write at Standard; gpt-6.1-sol shows the same five columns in the same table.
The only value wire_api accepts
The official config reference lists responses as the only supported value, and it is the default when omitted; a community write-up says setting chat throws an error.
Provider keys belong in user-level config
Codex ignores model_provider and model_providers when they appear in a project-local .codex/config.toml, so provider keys have to live in the user-level ~/.codex/config.toml.
What the snippet does: model picks the model, model_provider picks the endpoint
As of November 20, 2026, a custom provider setup in Codex has two layers. The top-level model key picks the model, and the official reference uses gpt-6.1-sol as its example value. The top-level model_provider key points at one table under model_providers, where name is the display name, base_url is the API base URL, env_key is the environment variable holding the API key, and wire_api is the protocol.
You end up editing it after upgrading to 0.171.0 because the catalog layer moved and Sol is now the default catalog selection. Keep three things apart: the catalog default decides what runs when model is absent, model decides what runs in this session, and model_provider decides where the request goes.
2026-11-20: Sol enters the default catalog, TUI workspace on by default
Two changes in this release touch this configuration directly, and both are quoted from the official release notes.
- 1Model catalogs: Added GPT-6.2 Sol and Astra to model catalogs, with Sol as the default catalog selection.
- 2Interface: Enabled the TUI workspace by default, with searchable document previews, image previews, and subagent transcripts alongside the main conversation. Community side: developers wrote on forums after 2026-11-17 that they reverted to 6.1 after a couple of hours of testing, and others said 6.2 Sol keeps claiming it is in plan mode.
Three dates: community first, one 0.170.0 change, then 0.171.0 ships
From 2026-11-17 the community started comparing GPT-6.2 Sol and Astra inside Codex, and the thread kept drawing comments through 2026-11-19 (community reports). To try it yourself, the official model page gives codex --model gpt-6.2-sol, with codex exec and -m gpt-6.2-sol for non-interactive runs and /model for switching inside an interactive session.
rust-v0.170.0 shipped on 2026-11-18, and one of its changes has OpenTelemetry HTTP exporters refresh their auth headers on their own instead of needing a manual reset. It is 0.171.0 that makes GPT-6.2 Sol the catalog default, so 0.170.0 does not change the default your config.toml selects against.
rust-v0.171.0 shipped on 2026-11-20: the catalogs gained GPT-6.2 Sol and Astra, Sol became the default catalog selection, and the TUI workspace turned on by default. This is the moment to set the top-level model key to gpt-6.2-sol for a custom provider.
What is confirmed vs community claims and unpublished details
Confirmed (Codex release notes, config reference, model page and pricing page, verbatim)
As of November 20, 2026, these points can be checked one by one.
- 1Release: rust-v0.171.0 shipped on 2026-11-20, added GPT-6.2 Sol and Astra to the model catalogs, made Sol the default catalog selection and enabled the TUI workspace by default.
- 2Model keys: model is the model to use, with
gpt-6.1-solas the reference example; model_provider is a provider id from model_providers and defaults to openai, while the built-in openai, ollama and lmstudio ids are reserved and cannot be overridden. - 3Provider table: base_url is the API base URL for the provider, env_key is the environment variable supplying the API key, name is the display name, and wire_api supports responses only and is the default when omitted.
- 4Scope: model_provider and model_providers are ignored in a project-local .codex/
config.toml, and the official fix is to put provider, notification, and telemetry keys in user-level config instead. - 5One-off runs: codex
--modelgpt-6.2-sol; for non-interactive runs codex exec with -mgpt-6.2-sol; inside an interactive session /model switches models. - 6Official rates:
gpt-6.2-solbills $2.00 input, $10.00 output, $0.10 cached input, $2.50 for a 30m cache write and $5.00 for a 12h cache write at Standard, and $1.00 input with $5.00 output at Batch;gpt-6.1-solshows the same five Standard columns.
Community claims and details the vendor has not published
None of the following is an official conclusion, so do not plan around it.
- 1"6.2 Sol must cost more than 6.1" — the official pricing page shows the same five Standard columns and the same two Batch columns for both; cost depends on how you use the model and on model behaviour, not on the label.
- 2"wire_api accepts chat too" — a leftover from older tutorials; a community write-up calls chat a hard error, and the official reference lists responses only.
- 3Developers in the community report reverting to 6.1 after a couple of hours of testing, and others report 6.2 Sol repeatedly claiming it is in plan mode; that is hands-on experience, not an official conclusion.
- 4Fast and Ultrafast rate tables and the later multiplier change follow the official pricing page as it stands on the day you check it.
Leave model out and follow the catalog default vs write model in config.toml
A: no model key, the catalog default decides
With no model key at the top level, Codex runs whatever the catalog default is, and since 0.171.0 that is GPT-6.2 Sol, so the same config.toml may switch to another model after an upgrade, with the rates and the feel changing with it. If you want to stay on 6.1, do not leave model selection to the default.
B: pin model = "gpt-6.2-sol"
Pinning model = "gpt-6.2-sol" at the top level keeps every upgrade on that line's choice, whatever the catalog default becomes; switch back to 6.1 by editing the same line to gpt-6.1-sol, and try Astra once with codex -m gpt-6.2-astra without touching the file.
Five steps: point gpt-6.2-sol at a third-party base_url
- 1Check the version first: run codex
--versionand expect 0.171.0, the rust-v0.171.0 release from 2026-11-20, whose catalog makes Sol the default. - 2Open the user-level config at
~/.codex/config.tomland leave the project-local .codex/config.tomlalone, because model_provider and model_providers are ignored there. - 3Add two top-level lines: model = "
gpt-6.2-sol" and model_provider = "qcode". - 4Right below them add the provider table [model_providers.qcode] with four lines: name = "qcode", base_url = "
https://api.qcode.cc/openai", wire_api = "responses" and env_key = "QCODE_API_KEY"; the base_url carries no /v1 suffix. - 5Export the key in the same shell with export
QCODE_API_KEY=yourcr_ key, then run codex--modelgpt-6.2-solon a real task; to stay on 6.1 change the model line togpt-6.1-sol, and to try Astra once run codex -mgpt-6.2-astra.
The QCode variant: one cr_ key, Codex on the openai endpoint
QCode is an independent API access service with no affiliation to OpenAI, and access is split by protocol.
- 1Codex CLI speaks Responses: set base_url to
https://api.qcode.cc/openaiwith no /v1 suffix, wire_api = "responses" and env_key = "QCODE_API_KEY"; in mainland China you can usehttps://asia.qcode.cc/openai, and in North America or Europehttps://us.qcode.cc/openai. - 2GPT models on offer include
gpt-6.2-sol,gpt-6.2-astra,gpt-6.1-solandgpt-6.1-terra, so switching models is a matter of changing the model ID. - 3The same cr_ key drives Claude models in Claude Code such as
claude-fable-5-2andclaude-sonnet-5-5throughANTHROPIC_BASE_URL=https://api.qcode.cc/api, while OpenAI-compatible clients reachqwen3.8-maxandglm-5.3throughhttps://api.qcode.cc/openai/v1. Billing is per token, with live per-model rates on /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; if you already have a Personal Plan, add a booster pack when you need more quota.
Common questions
What is the default model after 0.171.0?
GPT-6.2 Sol. The release notes say both models were added to the catalogs with Sol as the default catalog selection, and that default applies only when no model key sits at the top level; write model = "gpt-6.2-sol" or gpt-6.1-sol and your value wins. To see which one is active for a given session, run /model inside an interactive session.
Do 6.2 Sol and 6.1 Sol cost the same officially?
Yes, at OpenAI's official Standard tier: on the pricing page gpt-6.2-sol and gpt-6.1-sol both bill $2.00 input, $10.00 output, $0.10 cached input, $2.50 for a 30m cache write and $5.00 for a 12h cache write. Batch matches too, at $1.00 input and $5.00 output. The difference sits in model behaviour, not in those columns.
Can wire_api be set to chat?
No. The official config reference says it word for word: responses is the only supported value, and it is the default when omitted. A community write-up says the chat value throws an error, and the official reference lists no second value, so an old config that still sets wire_api = "chat" needs that value swapped for responses.
Why is the .codex/config.toml in my project ignored?
Because provider keys are not project-local: when model_provider or model_providers appear in a project-local .codex/config.toml, Codex ignores them, and the official guidance is to put provider, notification, and telemetry keys in user-level config instead, meaning ~/.codex/config.toml. Keep sandbox and approval settings in the project file.
Does base_url need a /v1 suffix?
Not for Codex: with wire_api = "responses" the base_url stops at https://api.qcode.cc/openai, and the reference only asks for the provider's API base URL, with no /v1 layer. The /v1 form belongs to OpenAI-compatible chat/completions clients such as the OpenAI SDK or Cline at https://api.qcode.cc/openai/v1, so keep the two endpoints apart.
How do I use Astra for a single run?
Run it once without touching the config: codex --model gpt-6.2-astra, or codex exec with -m gpt-6.2-astra for non-interactive work, and use /model inside an interactive session. The model page says Astra and Sol each support a context window of up to 1 million tokens, and they are priced differently: Astra is $8.00 input and $40.00 output at Standard, Sol is $2.00 and $10.00.
Sources
OpenAI Codex releases rust-v0.171.0 and rust-v0.170.0 (github.com/openai/codex/releases, 2026-11-20 and 2026-11-18); Codex config reference (developers.openai.com/codex/config-reference, checked 2026-11-20); Codex models (developers.openai.com/codex/models, checked 2026-11-20); OpenAI API pricing (developers.openai.com/api/docs/pricing, checked 2026-11-20); community threads on news.ycombinator.com from 2026-11-17 and still active on 2026-11-19, used as community context only. All checked on 2026-11-20.
The default moved, the snippet does not have to: one model line, one base_url
Create a QCode account for a cr_ key, set base_url=https://api.qcode.cc/openai and env_key=QCODE_API_KEY in ~/.codex/config.toml, and call multiple models billed per token; if you are not ready for a monthly plan, buy a pay-as-you-go pack first, with rates on /models.
Related reading
Pay-as-you-go Packs: no monthly plan, valid 7 days
Buy a one-time quota valid for 7 days or add a booster pack to a Personal Plan; per-model rates are on /models.
Run Claude Opus 5.5 and GPT-6.2 Sol Pay-As-You-Go: Claude Code and Codex CLI 0.171.0 Setup (2026-11)
As of 2026-11-22: use one cr_ key for claude-opus-5-5 in Claude Code (/api) and gpt-6.2-sol in Codex CLI 0.171.0 (/openai), with official rates and global payments.
GPT-6.2 Sol vs Claude Fable 5.2: Coding Cost
As of 2026-11-19: Sol $2.00 in / $10.00 out vs Fable 5.2 at $8 / $40; a sample session bills $3.15 vs $12.30, with Batch at 50% and the 272K surcharge.
Crypto payment guide
Nine coins, 56 networks, zero platform fee for AI coding API credits.
OpenAI's release notes, config reference, model page and pricing page were checked on 2026-11-20 and are authoritative at github.com/openai/codex and developers.openai.com. QCode.cc is an independent API access service with no affiliation to and no endorsement from OpenAI, Anthropic or other vendors. QCode model availability and live rates follow qcode.cc/models.