The GPT-5.3-Codex Countdown: Migrating Your Codex CLI Configuration Before the June 30 API Deadline

Updated

codex-cligpt-5.3-codexgpt-5.2model-deprecationmigrationconfig.tomlgpt-5.5pricing

The GPT-5.3-Codex Countdown: Migrating Your Codex CLI Configuration Before the June 30 API Deadline

On 12 June 2026, GPT-5.2 vanished from the ChatGPT model picker without advance notice — despite the official deprecations page listing an August 10 shutdown date 1 2. Existing conversations were silently migrated to GPT-5.5 equivalents. Three days later, GPT-5.3-Codex remains accessible via the API, but the clock is ticking: new API requests to all three retiring models — GPT-5.2, GPT-5.2-Codex, and GPT-5.3-Codex — will be blocked on 30 June 2026 3. That gives Codex CLI practitioners exactly fifteen days to audit their configurations, update their profiles, recalibrate their cost models, and verify their CI/CD pipelines still function.

This article provides a concrete migration checklist for teams still running deprecated model strings in config.toml, named profiles, or codex exec automation scripts.

What Is Actually Retiring

Three models are leaving the API on 30 June 2026. A fourth deadline follows shortly after.

gantt
    title GPT-5.x Model Retirement Timeline
    dateFormat YYYY-MM-DD
    axisFormat %d %b

    section ChatGPT Picker
    GPT-5.2 removed (no notice)  :done, cp1, 2026-06-12, 1d

    section API — New Requests Blocked
    GPT-5.2 / GPT-5.2-Codex / GPT-5.3-Codex :crit, api1, 2026-06-30, 1d

    section API — Legacy Snapshots
    All legacy codex model snapshots removed :api2, 2026-07-23, 1d

    section API — Chat Aliases
    gpt-5.2-chat-latest / gpt-5.3-chat-latest :api3, 2026-08-10, 1d

    section Full Shutdown
    All deprecated endpoints removed :api4, 2026-12-31, 1d
Model API Cutoff Recommended Replacement
gpt-5.2 30 June 2026 gpt-5.5
gpt-5.2-codex 30 June 2026 gpt-5.5
gpt-5.3-codex 30 June 2026 gpt-5.5 or gpt-5.4
gpt-5.2-chat-latest 10 August 2026 gpt-5.5
gpt-5.3-chat-latest 10 August 2026 gpt-5.5

The critical detail: the 30 June deadline blocks new API requests. Existing in-flight sessions may continue briefly, but no new codex exec runs, no new interactive sessions, and no new CI pipeline invocations will succeed against these model endpoints after that date 3.

Why the Early Removal Matters for CLI Users

When GPT-5.2 disappeared from ChatGPT on 12 June, users discovered the change through forum threads rather than official communications 1. The deprecations documentation still showed an August 10 date for the API alias gpt-5.2-chat-latest, creating confusion about which timeline applied to which surface 2.

For Codex CLI users authenticating via ChatGPT login (the default path for Plus and Pro subscribers), this distinction is critical. ChatGPT-authenticated sessions use the model picker, and that picker no longer offers GPT-5.2. If your config.toml hardcodes model = "gpt-5.2", the CLI will attempt to use it but the ChatGPT session will route to GPT-5.5 silently — or fail, depending on timing and backend rollout state 4.

API-key users have until 30 June for GPT-5.2 and GPT-5.3-Codex, but the prudent move is to migrate now rather than discover failures in production on the cutoff date.

The Cost Reality: GPT-5.3-Codex vs GPT-5.5

The migration is not cost-neutral. GPT-5.3-Codex was remarkably cheap for its capability tier.

Metric GPT-5.3-Codex GPT-5.5 GPT-5.4
Input (per 1M tokens) $0.75 $5.00 $2.50
Cached input (per 1M tokens) — $0.50 $0.25
Output (per 1M tokens) $4.00 $30.00 $15.00
Context window 200K 1M 200K

GPT-5.5 output tokens cost 7.5x more than GPT-5.3-Codex 5 6. Even accounting for GPT-5.5’s reported efficiency gains — fewer tokens to produce equivalent results 3 — the cost increase is substantial. For teams running heavy codex exec pipelines, the monthly bill will rise unless mitigated.

GPT-5.4 sits in the middle at $2.50/$15.00 and remains a strong choice for coding-focused workloads where the million-token context window is unnecessary 6.

The Fifteen-Day Migration Checklist

Days 1-3: Audit

1. Scan for deprecated model strings.

# Find all references to deprecated models across your codebase
grep -rn 'gpt-5\.2\|gpt-5\.3-codex' \
  ~/.codex/ \
  .codex/ \
  .github/workflows/ \
  AGENTS.md \
  --include='*.toml' --include='*.yaml' --include='*.yml' --include='*.md' --include='*.sh'

Check every layer of the configuration hierarchy: user-level (~/.codex/config.toml), project-level (.codex/config.toml), named profiles (~/.codex/<profile>.config.toml), and CI/CD pipeline scripts that invoke codex exec --model 7.

2. Run codex doctor to surface configuration warnings.

codex doctor --json | jq '.configuration'

Recent versions of codex doctor flag deprecated model references in the configuration report 8. If you see warnings about model availability, act on them.

3. Catalogue your CI/CD model references.

Search GitHub Actions workflows, GitLab CI files, and any scripts that call codex exec with a --model flag:

grep -rn 'codex exec.*--model.*gpt-5\.\(2\|3\)' .github/ ci/ scripts/

Days 4-7: Update

4. Update config.toml model references.

For most teams, the migration is a single-line change:

# Before
model = "gpt-5.3-codex"

# After — full capability replacement
model = "gpt-5.5"

# Or — cost-conscious alternative
model = "gpt-5.4"

If you use named profiles for different task types, update each profile file individually 7:

# ~/.codex/review.config.toml — code review profile
model = "gpt-5.5"
model_reasoning_effort = "high"

# ~/.codex/scaffold.config.toml — scaffolding profile
model = "gpt-5.4"
model_reasoning_effort = "medium"

5. Implement a cost-mitigation routing strategy.

The price jump from GPT-5.3-Codex to GPT-5.5 is steep. Use named profiles to route tasks to the cheapest capable model 9:

# ~/.codex/config.toml — default for interactive work
model = "gpt-5.5"

# ~/.codex/batch.config.toml — CI/CD and bulk operations
model = "gpt-5.4"
service_tier = "flex"

The service_tier = "flex" setting routes requests to off-peak capacity at reduced cost, suitable for non-urgent pipeline work 9.

6. Update CI/CD pipeline model references.

# .github/workflows/codex-review.yml
- name: Run Codex review
  run: codex exec --model gpt-5.5 --approval-mode full-auto "Review the PR diff"

For pipelines that previously relied on GPT-5.3-Codex’s lower cost per run, consider switching to GPT-5.4 with --profile batch to manage budget impact.

Days 8-12: Validate

7. Run integration tests against the new model.

Model swaps can subtly alter output characteristics. GPT-5.5 is more capable than GPT-5.3-Codex but may produce different formatting, different code style preferences, or different levels of verbosity 3. Run your standard development tasks and compare:

# Test interactive session
codex --model gpt-5.5 "Refactor the authentication module to use dependency injection"

# Test headless execution
codex exec --model gpt-5.5 --approval-mode full-auto \
  "Run the test suite and fix any failures" 2>&1 | tee migration-test.log

8. Verify hook and guardian compatibility.

If you use PreToolUse or PostToolUse hooks, confirm they work correctly with GPT-5.5’s output patterns. The model’s response structure uses the Responses API format, which is the same API surface GPT-5.3-Codex used in recent Codex CLI versions 10.

9. Check AGENTS.md model-specific instructions.

Some teams include model-specific guidance in AGENTS.md. Search for references to deprecated models and update them:

grep -rn 'gpt-5\.2\|gpt-5\.3-codex' **/AGENTS.md

Days 13-15: Harden

10. Set up fallback routing.

If you are concerned about GPT-5.5 availability during the transition period, configure a provider fallback:

# Primary model
model = "gpt-5.5"

# Fallback for rate-limited scenarios
[model_providers.backup]
model = "gpt-5.4"

11. Pin your Codex CLI version.

Ensure your team is running a Codex CLI version that includes the latest model catalogue. Versions v0.137.0 and later include full GPT-5.5 support and updated model picker entries 8:

codex --version
# Should be >= 0.137.0

12. Document the change for your team.

Update your team’s development setup guide with the new model defaults. If you use managed configuration bundles (cloud config), push the updated model settings through your admin console 11.

The GPT-5.4 Middle Path

Not every workload justifies GPT-5.5’s pricing. GPT-5.4 at $2.50/$15.00 per million tokens is 3.3x cheaper on input and 2x cheaper on output than GPT-5.5, whilst still delivering strong coding performance 6. The trade-off is a 200K context window versus GPT-5.5’s million-token capacity.

flowchart TD
    A[Task arrives] --> B{Needs >200K context?}
    B -->|Yes| C[Route to GPT-5.5]
    B -->|No| D{Cost-sensitive pipeline?}
    D -->|Yes| E[Route to GPT-5.4 with flex tier]
    D -->|No| F{Needs maximum reasoning?}
    F -->|Yes| C
    F -->|No| G[Route to GPT-5.4]

For teams migrating from GPT-5.3-Codex, GPT-5.4 is often the more accurate replacement in terms of cost profile, whilst GPT-5.5 is the more accurate replacement in terms of capability.

What Happens If You Do Not Migrate

After 30 June, API calls specifying gpt-5.3-codex, gpt-5.2-codex, or gpt-5.2 will receive HTTP 404 errors 3. Codex CLI sessions will fail at startup if the configured model is unavailable. CI/CD pipelines using codex exec --model gpt-5.3-codex will exit with non-zero status codes, breaking builds.

The December 31 full shutdown is the final backstop, but the practical deadline for most Codex CLI users is 30 June — fifteen days from today.

Citations

  1. OpenAI Developer Community, “GPT-5.2 disappeared from ChatGPT without advance notice, despite Aug 10 shutdown date in deprecations docs,” June 2026. https://community.openai.com/t/gpt-5-2-disappeared-from-chatgpt-without-advance-notice-despite-aug-10-shutdown-date-in-deprecations-docs/1383491 ↩ ↩2

  2. OpenAI, “Deprecations — OpenAI API,” June 2026. https://developers.openai.com/api/docs/deprecations ↩ ↩2

  3. ByteIota, “OpenAI Retires GPT-5.2 and GPT-5.3-Codex: Migrate Now,” June 2026. https://byteiota.com/openai-gpt52-gpt53-codex-sunset-migration/ ↩ ↩2 ↩3 ↩4 ↩5

  4. TechTimes, “OpenAI Retires GPT-5.2 and Moves Everyone to GPT-5.5: What Changes for ChatGPT Users and Developers,” 13 June 2026. https://www.techtimes.com/articles/318345/20260613/openai-retires-gpt-52-moves-everyone-gpt-55-what-changes-chatgpt-users-developers.htm ↩

  5. OpenRouter, “GPT-5.5 — API Pricing & Benchmarks,” June 2026. https://openrouter.ai/openai/gpt-5.5 ↩

  6. AI Pricing Guru, “OpenAI API Pricing 2026: GPT-5.5, GPT-5.4, o3 & All Models,” June 2026. https://www.aipricing.guru/openai-pricing/ ↩ ↩2 ↩3

  7. OpenAI, “Config basics — Codex,” June 2026. https://developers.openai.com/codex/config-basic ↩ ↩2

  8. OpenAI, “Changelog — Codex,” June 2026. https://developers.openai.com/codex/changelog ↩ ↩2

  9. Codex Knowledge Base, “Codex CLI Model Routing in May 2026: GPT-5.5, GPT-5.4, Codex-Spark, and When to Use Each,” May 2026. https://codex.danielvaughan.com/2026/05/07/codex-cli-model-routing-may-2026-gpt55-gpt54-spark-decision-framework/ ↩ ↩2

  10. OpenAI, “Features — Codex CLI,” June 2026. https://developers.openai.com/codex/cli/features ↩

  11. ChatGPT AI Hub, “OpenAI Sunsets GPT-5.2 and GPT-5.3-Codex: What Developers Need to Know About the Model Transition,” June 2026. https://chatgptaihub.com/openai-sunsets-gpt-5-2-gpt-5-3-codex-model-transition-guide/ ↩