Model Deprecation Checklist: Find and Replace Pinned Model IDs
Checked 2026-10-22. Copilot already removed Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5 on 2026-10-19. Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra are still scheduled for 2026-11-16. Those Copilot posts give display names only. Ark Coding Plan: minimax-m3 shuts down 2026-11-02 14:00 UTC+8; kimi-k2.7-code shuts down 2026-10-28 14:00 UTC+8. Includes rg commands.
On 2026-10-22 this guide reopened the Copilot models removed on 2026-10-19. Display names follow that post. It has no model IDs, and none are invented below. The 2026-11-16 Copilot post and the Codex notes were opened on 2026-10-17 and were not reopened. The DeepSeek alias note was added on 2026-10-18 after reopening the pricing page and the changelog. The 2026-10-23 table on the OpenAI deprecations page was copied on 2026-10-17. That check date has passed. The page was opened again on 2026-10-22, and that batch was still listed as a future shutdown. Whether it is actually off follows the deprecations page. The Ark rows were added on 2026-10-22 after opening the feature release and the model deprecation pages. If a page does not say it, this guide says unknown.
Copilot prices are not repeated here. See Copilot AI Credits billing. DeepSeek prices and IDs are in DeepSeek API model IDs.
Bottom line
Keep the lists separate.
- GitHub Copilot is two batches.
- Already removed, 2026-10-19: Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5. The suggested replacement for each one is in the table below. The post uses display names. It does not publish a model ID.
- Scheduled, 2026-11-16: Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra. The suggested replacements are Claude Opus 5.5, and GPT-6.1 Sol or GPT-6 Luna. Individual annual subscribers keep Claude Opus 4.8. That post also uses display names and does not publish a model ID you can search for.
- Codex already removed GPT-5.5 on 2026-10-14 at noon PDT (19:00 UTC) for users signed in with ChatGPT. The OpenAI API and API-key sessions are not affected. The note says to use
gpt-6.1-solwhen your plan and client still offer it. CLI 0.161.0 (published 2026-10-07) made GPT-6.1 Sol the default in the bundled catalog and the Amazon Bedrock catalogs. A line you wrote yourself,model = "gpt-5.5"inconfig.toml, wins over that default. Upgrading the CLI does not rewrite it. The Copilot display name GPT-5.5 is a different product. It does not replace this ID. - The OpenAI API deprecations page loaded on 2026-10-17. Cloudflare did not block it. It schedules 2027-01-14 shutdowns for specific GPT-4o and GPT-4.1 snapshots. A nearer batch is 2026-10-23. The table below was copied on 2026-10-17, and that check date has passed. Whether those IDs are off, and whether the replacements changed, follows the deprecations page. Undated aliases
gpt-4oandgpt-4.1do not have their own rows. Whether they shut down the same day is unknown. - DeepSeek:
deepseek-v4-flashanddeepseek-v4-flash-vision-expare now compatibility aliases. The pricing page says those models are retired and that requests are served by DeepSeek-V4.1-Flash-1017 at the Flash price. New code should senddeepseek-flash.deepseek-chatanddeepseek-reasonerwere scheduled to stop on 2026-07-24 and are not on the current pricing page. Whether they still respond is unknown. - Volcengine Ark Coding Plan:
minimax-m3shuts down at 2026-11-02 14:00:00 (UTC+8). The page suggestsdoubao-seed-evolvingorkimi-k3.kimi-k2.7-codeshuts down at 2026-10-28 14:00:00 (UTC+8). The page suggestskimi-k3. On the same page,minimax-latestandkimi-lateststill point at those old IDs.
Copilot is not the main door for every team. If you use Codex, or you send a model ID from your own service, the second and third items are the ones to search for. If the repo still says deepseek-v4-flash or deepseek-v4-flash-vision-exp, item 4 is the one to change to deepseek-flash, unless you mean to keep the compatibility alias. If an Ark config still has an ID from item 5, use that section. This guide does not say anything about network access.
What the pages actually say
Copilot: removed on 2026-10-19
Source: GitHub Copilot models deprecated on October 19, 2026 (dated October 19, 2026). Opened on 2026-10-22.
The sentence is “As of today, October 19, 2026, we have deprecated the following models.” This batch is already removed. The same post says these deprecations were announced on September 18, 2026. That September post was not opened for this check.
| Model in the post | Deprecation date | Suggested alternative |
|---|---|---|
| Gemini 3.7 Flash | 2026-10-19 | Gemini 3.8 Flash |
| GPT-5.5 | 2026-10-19 | GPT-6.1 Sol |
| GPT-5.4 | 2026-10-19 | GPT-6.1 Sol |
| GPT-5.4 mini | 2026-10-19 | GPT-6 Luna |
| GPT-5 mini | 2026-10-19 | GPT-6 Luna |
| Grok 4.5 | 2026-10-19 | Grok 4.7 |
The scope word is “most GitHub Copilot experiences.” The post names Copilot Chat, inline edits, ask mode, agent mode, and code completions. Experiences it does not name are unknown.
The post does not give a model ID for these six. Unknown. Do not invent a lowercase hyphenated ID. GPT-5.5 in this table is a display name. gpt-5.5 in the Codex note is an ID from another product.
Copilot Enterprise administrators may need to enable the alternatives in model policies under Copilot settings. The post says that after the policy is on, the model shows up in the Copilot Chat model selector in VS Code and on github.com. It also says no action is required to remove the deprecated models. Workflows and integrations that still select the old display name need the replacement from that row.
This batch is a different day from the three models dated 2026-11-16. November 16 has not arrived.
Copilot: scheduled for 2026-11-16
Source: Upcoming deprecation of selected GitHub Copilot models in mid-November (dated October 15, 2026). That post was opened on 2026-10-17 and was not reopened. These three are still before their deprecation date.
| Model in the post | Deprecation date | Suggested alternative |
|---|---|---|
| Claude Opus 4.8 | 2026-11-16 | Claude Opus 5.5 |
| Claude Opus 5 | 2026-11-16 | Claude Opus 5.5 |
| GPT-5.6 Terra | 2026-11-16 | GPT-6.1 Sol or GPT-6 Luna |
The scope is all GitHub Copilot experiences. The post names Copilot Chat, inline edits, ask mode, agent mode, and code completions.
One exception: Claude Opus 4.8 remains available to individual subscribers on annual plans so they keep an Opus model. Claude Opus 5 and GPT-5.6 Terra do not get that sentence.
For Copilot Enterprise and Copilot Business, under default model enablement, the suggested alternatives are turned on automatically unless an admin has turned off the global default or explicitly disabled the model. After deprecation, the post says you do not have to remove the old models from policy yourself. Workflows that still select them need to move before that date.
The string IDs these three use inside Copilot clients are not on the changelog page. Unknown. Do not treat Anthropic’s claude-opus-4-8 as a Copilot ID.
Codex: GPT-5.5 is already gone
Source: ChatGPT Learn changelog, entry “GPT-5.5 retirement in Codex”, dated 2026-10-14.
It took effect at noon PDT on 2026-10-14, which is 19:00 UTC. The retirement applies to Codex for users signed in with ChatGPT. The OpenAI API, and Codex sessions authenticated with an API key, are not affected.
The same entry says to use GPT-6.1 Sol, ID gpt-6.1-sol, when your plan and client offer it. Update workspace defaults, saved model settings, managed configurations, custom agents, scheduled tasks, and scripts that still select gpt-5.5. That entry does not give file paths for the last three. Those paths are unknown. Search for the literal string instead.
An earlier entry on the same page, “GPT-5.5 retires from ChatGPT, ChatGPT Work, and Codex on October 14”, says all plans, including consumer, Business, Enterprise, and Edu, and says the OpenAI API is not included. Use the 2026-10-14 “effective” paragraph when they differ: it pulls API-key sessions out of the retirement.
The CLI default is a separate line. Both the 0.161.0 changelog entry and the GitHub release rust-v0.161.0 say GPT-6.1 Sol is now the default model in the bundled and Amazon Bedrock catalogs. The release was published at 2026-10-07T15:58:45Z. That is a catalog default, not a migration of your config file.
API: the deprecations page loaded
Source: Deprecations. The tables below are the markdown opened on 2026-10-17. On 2026-10-22 the page was opened again only to see whether the 2026-10-23 batch was still listed as a future shutdown. The page was not copied again cell by cell.
The page says you can call GET /v1/models and read each model’s optional shutdown_date. A date is YYYY-MM-DD. If none has been announced, the field is null. Retrieve a model exposes it too. That is more reliable than searching blog posts for the word “deprecation”.
Rows that show up most often in coding repos are below. Speech, image, and Evals have more rows. Use the source page for those. This guide does not copy the whole page.
2026-10-23 (copied 2026-10-17; that check date has passed). On 2026-10-22 the deprecations page still listed this batch under future shutdowns, dated October 23, 2026. It was not under Past deprecations. On the shutdown date and after it, use the deprecations page. Do not treat the copy below as proof that the models are already off. The same visit showed a fine-tune row for ft-davinci-002, with replacement gpt-5.6-terra. The 2026-10-17 copy does not include that row. The other cells were not re-copied one by one.
| Shutdown | IDs in that cell | Replacement |
|---|---|---|
| 2026-10-23 | gpt-3.5-turbo-0125, gpt-3.5-turbo, gpt-3.5-turbo-completions | gpt-5.6-terra |
| 2026-10-23 | gpt-4-0613, gpt-4, gpt-4-0613-completions, gpt-4-completions | gpt-5.6-sol |
| 2026-10-23 | gpt-4-1106-preview | gpt-5.6-sol |
| 2026-10-23 | gpt-4-turbo, gpt-4-turbo-2024-04-09, gpt-4-turbo-completions | gpt-5.6-sol |
| 2026-10-23 | gpt-4.1-nano, gpt-4.1-nano-2025-04-14 | gpt-5.6-luna |
| 2026-10-23 | gpt-4o-2024-05-13 | gpt-5.6-sol |
| 2026-10-23 | o1-2024-12-17, o1 | gpt-5.6-sol |
| 2026-10-23 | o1-pro-2025-03-19, o1-pro | gpt-5.6-sol (reasoning.mode: pro) |
| 2026-10-23 | o3-mini-2025-01-31, o3-mini | gpt-5.6-sol |
| 2026-10-23 | ft-o4-mini-2025-04-16 | gpt-5.6-terra |
| 2026-10-23 | o4-mini-2025-04-16, o4-mini | gpt-5.6-terra |
The same day also shuts down fine-tunes ft-gpt-3.5-turbo, ft-gpt-4, ft-gpt-4.1-nano-2025-04-14, and ft-babbage-002, plus the image model gpt-image-1. The page’s replacement for gpt-image-1 is gpt-image-2.5-sunburst or gpt-image-2.5-flare.
2026-12-11: the replacement for gpt-5-2025-08-07 and o3-2025-04-16 is gpt-5.6-sol. The replacement for o3-pro-2025-06-10 and gpt-5-pro-2025-10-06 is also gpt-5.6-sol, with reasoning.mode: pro. The replacement for gpt-5-mini-2025-08-07 is gpt-5.6-terra. The replacement for gpt-5-nano-2025-08-07 is gpt-5.6-luna.
2027-01-14 (announced 2026-10-14):
| Shutdown | Wording on the page | Replacement |
|---|---|---|
| 2027-01-14 | GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20) | gpt-6-luna |
| 2027-01-14 | GPT-4o mini (gpt-4o-mini-2024-07-18) | gpt-6-luna |
| 2027-01-14 | GPT-4.1 (gpt-4.1-2025-04-14) | gpt-6-luna |
| 2027-01-14 | GPT-4.1 mini (gpt-4.1-mini-2025-04-14) | gpt-6-luna |
Fine-tuned versions of these models follow the same shutdown date as the base model.
2027-04-01: the replacement for gpt-5.3-codex and gpt-5.1 is gpt-6-sol. The replacement for gpt-5.4-nano is gpt-6-luna.
Two easy mix-ups:
- The shutdown table does not list undated
gpt-4oorgpt-4.1as their own rows. The Responses create enum, opened on 2026-10-17, still included those aliases andgpt-5.5. That page was not reopened. An ID still listed in the enum that day only means the reference page had not removed it. When the aliases shut down: unknown. - The API page lists
gpt-5.6-terraas a replacement for some older snapshots. The Copilot post lists GPT-5.6 Terra as retiring on 2026-11-16. Those are different products. Follow the API page for API calls. Follow the Copilot post for Copilot.
Volcengine Ark Coding Plan: two shutdowns that have not arrived
Source: model deprecation, opened 2026-10-22. The feature-release page opened the same day matches the start dates: minimax-m3 on 10.19, kimi-k2.7-code on 10.13.
| Model | Shutdown | Suggested move |
|---|---|---|
minimax-m3 | 2026-11-02 14:00:00 (UTC+8) | doubao-seed-evolving or kimi-k3 |
kimi-k2.7-code | 2026-10-28 14:00:00 (UTC+8) | kimi-k3 |
Both rows stop new users at 10:00:00 (UTC+8) on the start day. A new user is someone who bought Coding Plan and has not used that model.
The same page’s “method one” table still says minimax-latest → minimax-m3 and kimi-latest → kimi-k2.7-code. That is not the same sentence as the migration advice. Search the latest aliases too.
rg -n -e 'minimax-m3' -e 'kimi-k2\.7-code' -e 'minimax-latest' -e 'kimi-latest' .
The console was not opened. Whether the dropdown still lists them is unknown. The longer note is in the Ark Coding Plan guide.
Search the repo
The commands use ripgrep (rg). Run them from the repository root.
Default rg skips gitignored files. That hides .env, .env.local, .claude/settings.local.json, and .github/copilot/settings.local.json often enough to matter. Search twice: once with defaults, once with --no-ignore, and exclude dependency directories with --glob.
Literals have their own trap. Claude Opus 5 also matches the replacement Claude Opus 5.5. claude-opus-5 also matches claude-opus-5-5. Default rg has no lookahead, so the patterns below say “5 is not followed by ., -, or a digit”. That runs without --pcre2.
The October 19 six have display names only. The command below searches those names and does not add an ID the post never published. GPT-5.4 also matches GPT-5.4 mini. Both names are on the removed list. Grok 4.5 also shows up in Cursor’s weekly-pool docs, which are a different product. Read the path. GPT-5.5 here is the Copilot display name. A repo hit on gpt-5.5 still belongs with the Codex section.
rg -n --hidden \
--glob '!.git/**' \
--glob '!node_modules/**' \
--glob '!dist/**' \
--glob '!.next/**' \
-e 'Gemini 3\.7 Flash' \
-e 'GPT-5\.4 mini' \
-e 'GPT-5\.4' \
-e 'GPT-5 mini' \
-e 'GPT-5\.5' \
-e 'Grok 4\.5' \
.
Include the same six display names on the --no-ignore pass. Still do not add an ID you made up.
rg -n --hidden \
--glob '!.git/**' \
--glob '!node_modules/**' \
--glob '!dist/**' \
--glob '!.next/**' \
-e 'gpt-5\.5' \
-e 'GPT-5\.6 Terra' \
-e 'gpt-5\.6-terra' \
-e 'Claude Opus 4\.8' \
-e 'claude-opus-4-8' \
-e 'claude-opus-4\.8' \
-e 'Claude Opus 5([^.]|$)' \
-e 'claude-opus-5([^-.0-9]|$)' \
-e 'gpt-4o-2024-' \
-e 'gpt-4o-mini-2024-07-18' \
-e 'gpt-4\.1-2025-04-14' \
-e 'gpt-4\.1-mini-2025-04-14' \
-e 'gpt-4\.1-nano' \
-e 'gpt-4-turbo' \
-e 'gpt-4-0613' \
-e 'gpt-5\.3-codex' \
-e 'gpt-5-2025-08-07' \
-e 'o3-2025-04-16' \
-e 'o3-mini' \
.
Then the gitignored config files:
rg -n --hidden --no-ignore \
--glob '!.git/**' \
--glob '!node_modules/**' \
--glob '!dist/**' \
--glob '!.next/**' \
-e 'gpt-5\.5' \
-e 'gpt-5\.6-terra' \
-e 'claude-opus-4-8' \
-e 'Claude Opus 4\.8' \
-e 'Claude Opus 5([^.]|$)' \
-e 'gpt-4o-2024-' \
-e 'gpt-4\.1-2025-04-14' \
-e 'gpt-5\.3-codex' \
.
Machine-local config is outside the repo. Skip files that are not there:
for f in \
"$HOME/.codex/config.toml" \
"$HOME/.copilot/settings.json" \
"$HOME/.claude/settings.json" \
"$HOME/.cursor/cli-config.json"
do
if [ -f "$f" ]; then
rg -n \
-e 'gpt-5\.5' \
-e 'gpt-5\.6-terra' \
-e 'claude-opus-4-8' \
-e 'Claude Opus 4\.8' \
-e 'Claude Opus 5([^.]|$)' \
-e 'gpt-4o-2024-' \
-e 'gpt-4\.1-2025-04-14' \
-e 'gpt-5\.3-codex' \
"$f"
fi
done
Project overrides, paths relative to the repo:
for f in \
.codex/config.toml \
.claude/settings.json \
.claude/settings.local.json \
.github/copilot/settings.json \
.github/copilot/settings.local.json \
.github/allowed_models.txt \
.env \
.env.example \
.env.local
do
if [ -f "$f" ]; then
rg -n -e 'gpt-5\.5' -e 'claude-opus-4-8' -e 'gpt-4o-' -e 'gpt-4\.1-' -e 'gpt-5\.3-codex' "$f"
fi
done
CI is usually under .github/workflows/. The first repo search already covers YAML. This one is only scripts and the example env file:
rg -n --hidden --no-ignore \
--glob '!.git/**' \
--glob '!node_modules/**' \
-g '*.{yml,yaml,sh,py,js,mjs,ts,toml,json,env}' \
-e 'gpt-5\.5' \
-e 'gpt-4o-2024-' \
-e 'gpt-5\.3-codex' \
.github scripts .env.example
What a clean search still misses:
- An ID built at runtime. A string prefix
gpt-4o-plus a date variable never appears as the final ID. Search for the prefixgpt-4o-as a second pass. modelset from an environment variable keeps the ID in the deploy platform, not in git. Check CI variables, host environments, and the secret store. There is no single official variable name. This guide does not guess one.- Chat history, a model picked in an IDE window, and a default set in a cloud console are not in these files.
- Pointing
rgat all of~/.cursoralso searches caches and old sessions. A hit there is not a startup config. The loop above only openscli-config.json.
Bare gpt-4o also matches gpt-4o-2024-08-06 and gpt-4o-mini. To look for the alias only:
rg -n -e '(^|[^A-Za-z0-9-])gpt-4o([^A-Za-z0-9-]|$)' -e 'gpt-4\.1([^0-9-]|$)' .
Hits from that command still have an unknown shutdown date. Match them against the snapshot table first. Do not rewrite an alias unless you know the request really sends the undated string.
Where the model ID is stored
“Verified” means an official doc opened on 2026-10-17 names that location. Otherwise the cell says not verified.
| Tool | Location | Status |
|---|---|---|
| Copilot Chat | Model dropdown in VS Code, Visual Studio, JetBrains, Eclipse, and Xcode. In VS Code it is the CURRENT-MODEL menu at the bottom of the chat view. Changing the chat model does not change inline suggestions. Copilot Extensions may override the model you pick | Verified: Changing the AI model for GitHub Copilot Chat |
| Copilot CLI, user | model in ~/.copilot/settings.json. "auto" lets Copilot choose. /model and copilot config write it. COPILOT_HOME moves the directory | Verified: CLI configuration directory |
| Copilot CLI, repo | model in .github/copilot/settings.json, only in a trusted working directory. Personal override: .github/copilot/settings.local.json (keep it gitignored) | Verified, same page |
| Copilot CLI subagents | subagents.agents.<name>.model in settings.json | Verified, same page |
| Copilot CLI allowlist | .github/allowed_models.txt. One ID or glob per line. fallback: names the model used when the requested one is not allowed. This is an allowlist, not the default model | Verified, same page. The documented example uses gpt-6-astra and claude-sonnet-*. Neither the October 19 post nor the November 16 post publishes a Copilot model ID |
| Copilot enterprise managed settings | In the enterprise .github-private repo: copilot/managed-settings.json and copilot/teams/*.json. The key is top-level model. Older permissions.model is still read when the top-level key is absent | Verified: Getting started, Enterprise managed settings |
| Copilot managed file on the machine | macOS /Library/Application Support/GitHubCopilot/managed-settings.json; Windows %ProgramFiles%\GitHubCopilot\managed-settings.json; Linux /etc/github-copilot/managed-settings.json | Verified: Choosing how to deploy |
| Copilot org policy UI | Enable or disable models in the organization or enterprise settings. Not in your app repo, so rg will not see it | Verified that the UI exists: Configuring access to AI models. Follow that page for which control to click |
Copilot key name inside VS Code settings.json | The Copilot docs opened for this check do not name a github.copilot.chat.* key | Not verified |
| Codex CLI / IDE | model = "...". User file ~/.codex/config.toml, project .codex/config.toml (the project must be trusted), profile ~/.codex/<name>.config.toml, Unix system file /etc/codex/config.toml. Precedence: CLI --model / --config, then project config, then profile, then user config, then cloud-delivered defaults, then system config, then built-in defaults | Verified: Config basics. The example on that page that day was model = "gpt-6.1-sol" |
| Codex one-off override | codex --model <id>, or codex --config model='"<id>"' (the value is a TOML string) | Verified: Advanced Config |
| Codex, see the active model | /model switches. /status shows the active model | Verified: CLI reference. The examples on that page still say gpt-5.4-mini and gpt-5.5. Those are sample names, not the post-October-14 availability list |
| Codex custom agents and scheduled tasks, file path | Named in the retirement note, no path given | Not verified |
| Claude Code | model field. Highest precedence first: /model during the session, claude --model at startup, ANTHROPIC_MODEL, the model field in a settings file, then ANTHROPIC_DEFAULT_MODEL for new sessions. Pressing Enter in /model writes the choice into user settings | Verified: Model configuration |
| Claude Code settings files | User ~/.claude/settings.json; project .claude/settings.json; local project .claude/settings.local.json; organization managed-settings.json (also MDM and server-managed settings from claude.ai) | Verified: Settings |
| Claude Code aliases | opus, sonnet, and haiku move as versions change. To pin a version, use a full ID such as claude-opus-5-5 from that page, or a variable such as ANTHROPIC_DEFAULT_OPUS_MODEL | Verified, model configuration page |
| Claude Code subagents | A model key in subagent frontmatter is included in the retirement check. Which directory holds that file | The field is verified. The directory is not |
| Cursor editor | Model picker at the top of the chat input. Default model: Cursor Settings > Models | Verified: Prompting agents. Which file stores that IDE choice: not verified |
| Cursor CLI | --model <id>. Persisted in the model field of ~/.cursor/cli-config.json (Windows: %USERPROFILE%\.cursor\cli-config.json). The field type is object. Project file .cursor/cli.json is documented as permissions only. /model selects inside a session | Verified: CLI configuration, parameters. Keys inside the model object: not verified. The page has no sample JSON for it |
| Cursor Cloud Agents | Dashboard setting Default model, used when a run does not specify one | Verified: Cloud Agents settings |
| OpenAI API | Request body field model on Responses. The enum that day still included gpt-5.5, gpt-4o, gpt-4.1, and dated snapshots | Verified: Create a response |
| OpenAI Chat Completions reference | Requesting that page on 2026-10-17 returned 403 | Not verified |
| Anthropic Messages API | Request field model. The enum that day included claude-opus-4-8, claude-opus-5, and claude-opus-5-5 | Verified: Messages. That spelling is the Anthropic API, not an ID from the Copilot post |
Official SDKs put that model field on the request. Search for the assignment. Do not search for only one client method name. Chat Completions method names were not checked.
One concrete case. After Codex 0.161.0, the built-in default is GPT-6.1 Sol. If ~/.codex/config.toml still has:
model = "gpt-5.5"
a ChatGPT-signed-in session still asks for the retired ID. Delete the line to fall through to the new catalog default, or set an ID your plan still offers. The config page’s example that day was gpt-6.1-sol. The retirement note says “when available”. Do not assume every plan can select it.
Check after you change it
Split the check by product. A “write a short bio” prompt is not the regression.
- Run the
rgcommands again. Old IDs should be gone from config and scripts. Strings left in articles, changelogs, or lockfiles need a look at the path before you delete them. - Codex (ChatGPT sign-in): start it and run
/status. The active model should be the one you just wrote. Then run a small task you already trust, such as “run the existing unit tests, do not edit code”. If/modelstill lists GPT-5.5, trust the screen and write down the client version. This guide was not checked by opening a model picker in the app. - API: send one request to the endpoint you actually use, with the replacement ID from the deprecations page, such as
gpt-6-lunaorgpt-6-sol. On failure, read the model name in the error body. This guide did not verify the current error code string. Do not memorize anerror.code. Then run the test or CI job that owns that call. The check is whether the test still passes, not whether two paragraphs match. - Copilot: the six names from October 19 are already removed. Open the chat model menu and see whether Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5 are still there. On Enterprise, confirm the replacement model’s policy is enabled first. The post says the selector in VS Code and on github.com shows it after that. The three names dated November 16 are still before that day. What you can check now is that workflows and managed settings already name the replacements from that post, and that Enterprise or Business has not disabled those replacements. After November 16, look for those three display names. An individual annual subscriber who still sees Claude Opus 4.8 matches the November post. This guide was not checked by opening a model picker in the app.
- Cursor Cloud Agents and Copilot org policy live in a web console. After the change, start one run that does not pin a model and read the model name on the task. rg does not cover that.
If GET /v1/models already shows a shutdown_date for the model you care about, put that date on your own calendar. Do not wait for the request to fail on the shutdown day.
Frequently Asked Questions
Do API-key Codex sessions also need gpt-5.5 removed?
If rg finds nothing, is the repo safe?
Do individual annual Copilot subscribers still change Opus 4.8?
Which model ID should a repo search use for the six models removed on October 19?
📚 Keep Reading
📚 Related Guides
AI Coding Agents in 2026: A Practical Roadmap from Autocomplete to Cloud Teammates
As of June 2026, AI coding tools are moving from autocomplete and AI IDEs into engineered agent workflows. This guide maps the shift across Codex, Claude Code, GitHub Copilot, Gemini CLI, MCP, security, and team adoption.
Volcengine Ark Coding Plan Guide: Tiers, Agent Plan Differences, and Setup Notes
Based on the official Volcengine Ark Agent/Coding Plan export: Coding Plan is a two-tier coding-focused subscription for individual developers, while Agent Plan is the upgraded multimodal + Harness package. Covers the boundary, model/activity changes, Base URL/API key rules, and setup notes for Cline, Claude Code, OpenCode, and Codex CLI.
China Coding API Roundup 2026: Ark vs Bailian vs MiniMax vs Zhipu vs DeepSeek
The most comprehensive 2026 comparison of Chinese AI coding API subscriptions: Volcengine Ark Lite ¥40/mo (¥9.9 first month), Alibaba Bailian ¥200, MiniMax ¥29, Zhipu GLM ¥49+, and DeepSeek pay-as-you-go. One big table, six scenarios, and a decision tree to find your fit in 30 seconds.
📖 Related Comparisons
AI Coding Tool Monthly Cost Comparison 2026: Copilot, Cursor, Codex, Claude Code, and China Coding Plans
Compare monthly fees, quota models, China accessibility, and overage risk across GitHub Copilot, Cursor, OpenAI Codex, Claude Code, Volcengine Ark, MiniMax, Alibaba Bailian, and other AI coding setups.
AI Code Review Tools Comparison 2026: Which One Is Worth It?
A comprehensive comparison of 6 mainstream AI code review tools in 2026: GitHub Copilot Code Review, CodeRabbit, Qodo Merge, Greptile, Claude Code, and Amazon Q Developer — evaluated on comment quality, false positive rate, China accessibility, and pricing.
AI Coding Subscription Comparison 2026: Claude, Codex, and Domestic Coding Plans
2026 AI coding subscriptions have entered the 5x/20x era: Claude Pro/Max and Codex Plus/Pro are the two flagship international options, while domestic coding plans (GLM/Ark/Bailian) compete on low price and direct China access. Using official data verified 2026-08-19, this three-way comparison covers price, quota mechanics, models, ecosystem, and China accessibility - with a decision checklist you can act on directly.
🔧 Tools in This Article
💰 Related money-saving plans
Based on the tools and APIs in this guide — get started for less.
Claude Code + Volcano Ark Agent Plan
One-Stop Agent ToolchainWire Claude Code into Volcano Ark Agent Plan (launched 2026-05-11): Doubao-Seed for coding, Seedance/Seedream for multimodal generation, plus built-in web search / memory / Auto routing Harness tooling — all China-native. Billed in AFP credits; Medium tier (¥200) and above include a 24/7 AI companion.
For: China-native developers needing multimodal + Agent toolchain · Users wanting Doubao Seedance / Seedream multimodal models
Cline + Volcengine Ark Coding Plan
Best ValueA high-value Agent setup. Cline is free and open-source with Plan/Act dual mode, connected to Volcengine Ark API for direct China access. Ark offers Lite and Pro tiers; since March 13, 2026, new-user subscriptions open in limited quantities each day at 10:30, with models including Doubao-Seed-Code/GLM-4.7/DeepSeek-V3.2/Kimi-K2-Thinking.
For: Budget-conscious developers in China · Users who want free model choice
🎁 Grab deals & free trials
Free quotas, first-month discounts, and sign-up links for tools & coding plans — kept up to date.
View all deals →