The June–July 2026 Model Sunset: Migrating from GPT-5.2 and GPT-5.2-Codex in Your Codex CLI Workflows

Updated

modelsmigrationdeprecationgpt-5.2gpt-5.5configci-cdenterprise

The June–July 2026 Model Sunset: Migrating from GPT-5.2 and GPT-5.2-Codex in Your Codex CLI Workflows

OpenAI’s fleet simplification continues. GPT-5.2 was retired from the Codex web platform on 2 June 20261, GitHub Copilot drops it on 5 June2, and the API shutdown follows on 23 July for gpt-5.2-codex and 10 August for gpt-5.2-chat-latest3. If your config.toml, CI pipelines, or team profiles still reference these model identifiers, you have weeks — not months — to migrate.

This article is a practical checklist for Codex CLI users who need to audit, test, and cut over before the shutdowns bite.

What Is Being Retired and When

The retirement rolls out across three surfaces on different dates:

gantt
    title GPT-5.2 Family Retirement Timeline
    dateFormat YYYY-MM-DD
    axisFormat %d %b

    section Codex Platform
    GPT-5.2 removed (ChatGPT-auth users) :milestone, 2026-06-02, 0d

    section GitHub Copilot
    GPT-5.2 & GPT-5.2-Codex removed :milestone, 2026-06-05, 0d

    section OpenAI API
    gpt-5.2-codex shutdown         :crit, 2026-07-23, 1d
    gpt-5.2-chat-latest shutdown   :crit, 2026-08-10, 1d
Model ID Deprecated API Shutdown Replacement
gpt-5.2 8 May 2026 10 Aug 2026 gpt-5.5
gpt-5.2-chat-latest 8 May 2026 10 Aug 2026 gpt-5.5
gpt-5.2-codex 8 May 2026 23 Jul 2026 gpt-5.5
gpt-5.3-chat-latest 8 May 2026 10 Aug 2026 gpt-5.5

Sources: OpenAI API deprecations page3, GitHub Changelog2, OpenAI Codex platform announcement1.

Three important caveats:

  1. ChatGPT-auth CLI users lost access to GPT-5.2 in the TUI model selector on 2 June. If you authenticate via codex login with a ChatGPT account, the model is already gone1.
  2. API-key users retain access until the API shutdown dates above. Pipelines using OPENAI_API_KEY will keep working — but the clock is ticking3.
  3. GPT-5.3-Codex-Spark is not affected — it remains available and is not part of this retirement phase1.

Why GPT-5.2 Is Going Away

OpenAI cited fleet simplification: GPT-5.2 accounts for less than 1% of production usage, and maintaining older models in the active compute fleet carries overhead1. With GPT-5.5 scoring 73.1% on Expert-SWE versus GPT-5.2’s lower baseline, and using 72% fewer output tokens on equivalent tasks4, the case for keeping the older model no longer holds.

Step 1: Audit Your Configuration

Find Hardcoded Model References

Every Codex CLI configuration surface can pin a model. Check them all:

# User-level config
grep -n 'gpt-5\.[23]' ~/.codex/config.toml 2>/dev/null

# Project-level config (trusted repos)
find . -path '*/.codex/config.toml' -exec grep -ln 'gpt-5\.[23]' {} +

# Named profiles
find "${CODEX_HOME:-$HOME/.codex}" -name '*.config.toml' \
  -exec grep -ln 'gpt-5\.[23]' {} +

# AGENTS.md files (some teams embed model hints)
grep -rn 'gpt-5\.[23]' . --include='AGENTS.md' --include='AGENTS.override.md'

Check CI/CD Pipelines

If your pipelines use codex exec with an explicit --model flag, those invocations will break after the API shutdown:

# GitHub Actions workflows
grep -rn 'gpt-5\.[23]' .github/workflows/ 2>/dev/null

# GitLab CI
grep -n 'gpt-5\.[23]' .gitlab-ci.yml 2>/dev/null

# Makefiles and shell scripts
grep -rn 'gpt-5\.[23]' scripts/ Makefile 2>/dev/null

Check Environment Variables

Some teams set CODEX_MODEL or pass model IDs through environment variables:

env | grep -i 'model.*gpt-5\.[23]'
grep -rn 'CODEX_MODEL.*gpt-5\.[23]' .env* docker-compose*.yml 2>/dev/null

Step 2: Choose Your Replacement

The model landscape as of June 2026 offers several migration paths5:

Use Case Recommended Model Rationale
General development gpt-5.5 Default; best Expert-SWE score (73.1%); 72% fewer output tokens4
Budget-conscious teams gpt-5.4-mini Optimised for responsive coding and subagents5
Deep reasoning tasks gpt-5.4 Strong reasoning + tool use; absorbed GPT-5.3-Codex capabilities6
Near-instant iteration gpt-5.3-codex-spark Text-only; fastest latency; ChatGPT Pro only5

For most teams, gpt-5.5 is the straightforward replacement. It is the current default for new Codex CLI installations5 and handles the broadest range of tasks.

Step 3: Update config.toml

Simple Migration

Replace the model identifier in your user-level configuration:

# ~/.codex/config.toml

# Before
model = "gpt-5.2"

# After
model = "gpt-5.5"

Profile-Based Migration

If your team uses named profiles, update each one. This is the recommended approach for gradual rollout — you can migrate one profile at a time and compare results7:

# ~/.codex/config.toml

[profiles.fast]
model = "gpt-5.4-mini"
model_reasoning_effort = "medium"

[profiles.deep]
model = "gpt-5.5"
model_reasoning_effort = "xhigh"

[profiles.ci]
model = "gpt-5.5"
model_reasoning_effort = "medium"

Switch profiles with --profile:

codex --profile ci exec "Run the test suite and fix failures"

Avoid Pinning chat-latest Pointers

The -chat-latest model pointers (gpt-5.2-chat-latest, gpt-5.3-chat-latest) are snapshot aliases that do not auto-advance to newer models3. They are being retired alongside the base models. Prefer unversioned model names like gpt-5.5 which will receive updates:

# Avoid
model = "gpt-5.2-chat-latest"

# Prefer
model = "gpt-5.5"

Step 4: Test Before Cutting Over

Verify Model Availability

Check which models your account can access:

codex models

If you are authenticated via a ChatGPT account, GPT-5.2 will no longer appear in the list as of 2 June1.

Run a Side-by-Side Comparison

Use profiles to run the same task with both old and new models, then compare:

# Run with the old model (API-key users only, until shutdown)
codex --profile old exec "Refactor the authentication module" \
  --json -o /tmp/result-old.jsonl

# Run with the new model
codex --profile new exec "Refactor the authentication module" \
  --json -o /tmp/result-new.jsonl

Check Reasoning Effort Tuning

GPT-5.5 supports the full range of reasoning effort levels5. If you were using GPT-5.2 at high reasoning effort, you may find that medium on GPT-5.5 delivers comparable quality at lower cost and latency:

model = "gpt-5.5"
model_reasoning_effort = "medium"  # often sufficient for interactive work

Reserve xhigh for tasks that would “eat an afternoon of senior developer time”8.

Step 5: Update CI/CD Pipelines

GitHub Actions with codex-action

Update the model parameter in your workflow files:

# .github/workflows/codex-review.yml
- uses: openai/codex-action@v1
  with:
    model: "gpt-5.5"
    codex_args: "--approval-mode auto-edit"
  env:
    OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}

Non-Interactive Scripts

For codex exec invocations, update the --model flag or remove it to inherit the default:

# Before
codex exec --model gpt-5.2-codex "Generate the changelog" -o changelog.md

# After (explicit)
codex exec --model gpt-5.5 "Generate the changelog" -o changelog.md

# After (inherit default from config.toml)
codex exec "Generate the changelog" -o changelog.md

Validate with codex doctor

After updating configuration, run diagnostics to confirm the setup is coherent:

codex doctor --summary

The output should show your new model in the configuration section without warnings about deprecated model identifiers9.

Step 6: Enterprise Considerations

Managed Configuration (requirements.toml)

Enterprise administrators using cloud-managed FeatureRequirementsToml should update the enforced model list to exclude GPT-5.2 variants. Teams inheriting configuration from the organisation layer will automatically pick up the change7.

Azure OpenAI and Amazon Bedrock Users

If your model_providers table routes through Azure OpenAI or Amazon Bedrock, check the provider’s own deprecation schedule. Azure follows a separate timeline managed through the Foundry lifecycle policy10. Bedrock availability depends on the Mantle Responses API model catalogue and may not mirror OpenAI’s direct schedule.

GitHub Copilot Overlap

Teams using both Codex CLI and GitHub Copilot should note that GPT-5.3-Codex became the default for Copilot Business and Enterprise on 17 May 202611. Copilot Enterprise administrators need to enable alternative models through their Copilot settings before 5 June2.

Migration Decision Flowchart

flowchart TD
    A[Does config.toml reference gpt-5.2*?] -->|Yes| B[Are you using API-key auth?]
    A -->|No| C[No action needed]
    B -->|Yes| D[Update to gpt-5.5 before 23 July]
    B -->|No, ChatGPT auth| E[GPT-5.2 already removed - update now]
    D --> F[Test with codex doctor + side-by-side runs]
    E --> F
    F --> G[Update CI/CD pipelines]
    G --> H[Update team profiles and requirements.toml]
    H --> I[Monitor token usage - expect ~72% reduction]

Key Dates to Remember

  • 2 June 2026: GPT-5.2 removed from Codex platform (ChatGPT-auth users)1
  • 5 June 2026: GPT-5.2 and GPT-5.2-Codex removed from GitHub Copilot2
  • 23 July 2026: gpt-5.2-codex API shutdown3
  • 10 August 2026: gpt-5.2-chat-latest and gpt-5.3-chat-latest API shutdown3

The migration is straightforward for most teams: replace gpt-5.2 with gpt-5.5 in your config.toml and CI scripts, verify with codex doctor, and enjoy the 72% output token reduction. The only real risk is doing nothing and discovering your pipelines have broken on a Monday morning.

Citations

  1. OpenAI, “Retiring GPT-5.2 and GPT-5.3-Codex from Codex,” OpenAI Platform Announcement, June 2026. https://www.kucoin.com/news/flash/openai-to-discontinue-gpt-5-2-and-gpt-5-3-codex-in-codex-on-june-2 ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7

  2. GitHub, “Upcoming deprecation of GPT-5.2 and GPT-5.2-Codex,” GitHub Changelog, 1 May 2026 (updated 29 May 2026). https://github.blog/changelog/2026-05-01-upcoming-deprecation-of-gpt-5-2-and-gpt-5-2-codex/ ↩ ↩2 ↩3 ↩4

  3. OpenAI, “API Deprecations,” OpenAI Developers Documentation, accessed 1 June 2026. https://developers.openai.com/api/docs/deprecations ↩ ↩2 ↩3 ↩4 ↩5 ↩6

  4. OpenAI, “Introducing GPT-5.5,” OpenAI Blog, 23 April 2026. https://openai.com/index/introducing-gpt-5-5/ ↩ ↩2

  5. OpenAI, “Models — Codex,” OpenAI Developers Documentation, accessed 1 June 2026. https://developers.openai.com/codex/models ↩ ↩2 ↩3 ↩4 ↩5

  6. NxCode, “GPT 5.4 Complete Guide 2026: Features, Pricing, Benchmarks & How to Use,” NxCode, 2026. https://www.nxcode.io/resources/news/gpt-5-4-complete-guide-features-pricing-models-2026 ↩

  7. OpenAI, “Advanced Configuration — Codex,” OpenAI Developers Documentation, accessed 1 June 2026. https://developers.openai.com/codex/config-advanced ↩ ↩2

  8. OpenAI, “Codex Prompting Guide,” OpenAI Cookbook, 2026. https://developers.openai.com/cookbook/examples/gpt-5/codex_prompting_guide ↩

  9. OpenAI, “Codex CLI Changelog — v0.135.0,” OpenAI Developers, 28 May 2026. https://developers.openai.com/codex/changelog ↩

  10. Microsoft, “Azure OpenAI in Microsoft Foundry Model Retirements,” Microsoft Learn, 2026. https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirements ↩

  11. GitHub, “GPT-5.3-Codex is now the base model for Copilot Business and Enterprise,” GitHub Changelog, 17 May 2026. https://github.blog/changelog/2026-05-17-gpt-5-3-codex-is-now-the-base-model-for-copilot-business-and-enterprise/ ↩