Add first-class gpt-5.4-mini support#83
Merged
ndycode merged 6 commits intondycode:mainfrom Mar 18, 2026
Merged
Conversation
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
Contributor
📝 WalkthroughWalkthroughAdds GPT-5.4 Mini as a first-class model family (base + five effort-level variants) across configs, model normalization, prompt/cache mappings, reasoning-config logic, and tests; exports a new getReasoningConfig utility for deriving per-model reasoning settings. (≤50 words) Changes
Sequence Diagram(s)sequenceDiagram
participant Client
participant Transformer as RequestTransformer
participant ModelMap
participant Prompts as PromptManager
participant Config as ReasoningConfigStore
Client->>Transformer: send model id (e.g., openai/gpt-5.4-mini-high)
Transformer->>ModelMap: normalizeModel(model id)
ModelMap-->>Transformer: normalized id (gpt-5.4-mini-high)
Transformer->>Config: getReasoningConfig(normalized id, userConfig)
Config-->>Transformer: reasoning settings (effort, summary)
Transformer->>Prompts: selectPromptFamily(normalized base)
Prompts-->>Transformer: prompt template & cache key
Transformer-->>Client: transformed request (model, reasoning headers, prompt)
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
📝 Coding Plan
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
There was a problem hiding this comment.
Pull request overview
Adds first-class gpt-5.4-mini support across model normalization, model-family routing (Codex prompts + caches), configuration templates, and docs, with expanded regression coverage to keep legacy gpt-5-mini behavior intact.
Changes:
- Extend
normalizeModel,MODEL_MAP, and reasoning-effort rules to treatgpt-5.4-minias a distinct GPT‑5.4 family member. - Route
gpt-5.4-minithrough Codex prompt selection with an isolated cache key/file. - Add/update tests, docs, and shipped config templates for the new model family.
Reviewed changes
Copilot reviewed 11 out of 11 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| test/request-transformer.test.ts | Adds normalization/config/reasoning/transformRequestBody regression tests for gpt-5.4-mini. |
| test/model-map.test.ts | Asserts MODEL_MAP, getNormalizedModel, and isKnownModel include gpt-5.4-mini. |
| test/gpt54-models.test.ts | Adds broad normalization/family/reasoning tests for gpt-5.4-mini. |
| test/codex.test.ts | Verifies Codex getModelFamily recognizes gpt-5.4-mini distinctly. |
| test/codex-prompts.test.ts | Verifies prompt routing + cache isolation for gpt-5.4-mini. |
| lib/request/request-transformer.ts | Implements gpt-5.4-mini normalization + first-class reasoning behavior. |
| lib/request/helpers/model-map.ts | Adds gpt-5.4-mini keys and dated alias expansion to MODEL_MAP. |
| lib/prompts/codex.ts | Extends model families and cache/prompt files for gpt-5.4-mini. |
| docs/configuration.md | Documents reasoning-effort support for gpt-5.4-mini + clarifies legacy aliases. |
| config/opencode-modern.json | Adds gpt-5.4-mini as a first-class model with reasoning variants. |
| config/opencode-legacy.json | Adds legacy preset-style entries for gpt-5.4-mini-* efforts. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+99
to
+108
| // ============================================================================ | ||
| // GPT-5.4 Mini Models (latest efficient family) | ||
| // ============================================================================ | ||
| "gpt-5.4-mini": "gpt-5.4-mini", | ||
| "gpt-5.4-mini-none": "gpt-5.4-mini", | ||
| "gpt-5.4-mini-low": "gpt-5.4-mini", | ||
| "gpt-5.4-mini-medium": "gpt-5.4-mini", | ||
| "gpt-5.4-mini-high": "gpt-5.4-mini", | ||
| "gpt-5.4-mini-xhigh": "gpt-5.4-mini", | ||
| ...expandDatedAliases(`gpt-5.4-mini-${GPT_54_SNAPSHOT_DATE}`, "gpt-5.4-mini"), |
Comment on lines
+128
to
+147
| describe("GPT-5.4 Mini Model Normalization", () => { | ||
| it("should normalize gpt-5.4-mini base model", () => { | ||
| expect(normalizeModel("gpt-5.4-mini")).toBe("gpt-5.4-mini"); | ||
| expect(getNormalizedModel("gpt-5.4-mini")).toBe("gpt-5.4-mini"); | ||
| }); | ||
|
|
||
| it("should normalize all gpt-5.4-mini reasoning effort variants", () => { | ||
| const variants = [ | ||
| "gpt-5.4-mini-none", | ||
| "gpt-5.4-mini-low", | ||
| "gpt-5.4-mini-medium", | ||
| "gpt-5.4-mini-high", | ||
| "gpt-5.4-mini-xhigh", | ||
| ]; | ||
|
|
||
| for (const variant of variants) { | ||
| expect(normalizeModel(variant)).toBe("gpt-5.4-mini"); | ||
| expect(getNormalizedModel(variant)).toBe("gpt-5.4-mini"); | ||
| } | ||
| }); |
test/request-transformer.test.ts
Outdated
Comment on lines
+119
to
+128
| it('should normalize gpt-5.4-mini presets as a first-class family', async () => { | ||
| expect(normalizeModel('gpt-5.4-mini')).toBe('gpt-5.4-mini'); | ||
| expect(normalizeModel('gpt-5.4-mini-none')).toBe('gpt-5.4-mini'); | ||
| expect(normalizeModel('gpt-5.4-mini-low')).toBe('gpt-5.4-mini'); | ||
| expect(normalizeModel('gpt-5.4-mini-medium')).toBe('gpt-5.4-mini'); | ||
| expect(normalizeModel('gpt-5.4-mini-high')).toBe('gpt-5.4-mini'); | ||
| expect(normalizeModel('gpt-5.4-mini-xhigh')).toBe('gpt-5.4-mini'); | ||
| expect(normalizeModel('openai/gpt-5.4-mini-xhigh')).toBe('gpt-5.4-mini'); | ||
| expect(normalizeModel('gpt-5.4-mini-2026-03-05-high')).toBe('gpt-5.4-mini'); | ||
| }); |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
lib/request/request-transformer.ts (1)
87-90: Minor regex inconsistency with Pro pattern.The outer non-capturing group
(?:...)is unnecessary here. The Pro regex at line 83 doesn't have this wrapper:
- Pro:
/\bgpt(?:-| )5\.4(?:-| )pro(?:\b|[- ])/- Mini:
/(?:\bgpt(?:-| )5\.4(?:-| )mini(?:\b|[- ]))/Functionally equivalent, but could align for consistency.
🛠️ Optional: Remove unnecessary wrapper
- if (/(?:\bgpt(?:-| )5\.4(?:-| )mini(?:\b|[- ]))/.test(normalized)) { + if (/\bgpt(?:-| )5\.4(?:-| )mini(?:\b|[- ])/.test(normalized)) {🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@lib/request/request-transformer.ts` around lines 87 - 90, The regex for detecting "gpt-5.4-mini" uses an unnecessary outer non-capturing group — update the pattern used where you return "gpt-5.4-mini" to match the style of the Pro pattern (remove the outer (?: ... ) wrapper) so both regexes are consistent while keeping the inner structure \bgpt(?:-| )5\.4(?:-| )mini(?:\b|[- ]) intact.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@docs/configuration.md`:
- Around line 55-56: Update the documentation lines that claim legacy aliases
map only to base models and that only base/pro use context=1000000: state that
legacy `gpt-5`, `gpt-5-mini`, `gpt-5-nano` aliases all map to `gpt-5.4` (and
corresponding mini snapshots map to `gpt-5.4-mini` as supported in code), ensure
snapshot patterns `gpt-5.4-2026-03-05*` and `gpt-5.4-pro-2026-03-05*` mention
their mini equivalents (e.g., `gpt-5.4-mini-2026-03-05*`) map to their stable
forms, and change the note about context limits to reflect that shipped mini
templates also use context=1000000 (not just base/pro).
---
Nitpick comments:
In `@lib/request/request-transformer.ts`:
- Around line 87-90: The regex for detecting "gpt-5.4-mini" uses an unnecessary
outer non-capturing group — update the pattern used where you return
"gpt-5.4-mini" to match the style of the Pro pattern (remove the outer (?: ... )
wrapper) so both regexes are consistent while keeping the inner structure
\bgpt(?:-| )5\.4(?:-| )mini(?:\b|[- ]) intact.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 37c9ab73-7bfa-4f76-8149-b6e29ea2f38b
📒 Files selected for processing (11)
config/opencode-legacy.jsonconfig/opencode-modern.jsondocs/configuration.mdlib/prompts/codex.tslib/request/helpers/model-map.tslib/request/request-transformer.tstest/codex-prompts.test.tstest/codex.test.tstest/gpt54-models.test.tstest/model-map.test.tstest/request-transformer.test.ts
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
lib/request/request-transformer.ts (1)
480-514:⚠️ Potential issue | 🟡 MinorUse token-aware
minidetection here.These substring checks also match the
-minimalsuffix that this file already accepts as a reasoning variant. That makesgetReasoningConfig("gpt-5.4-minimal")look like first-class Mini, whilegpt-5-minimal/gpt-5.1-minimalfall into the lightweight-alias path.normalizeModeldoes not classify those names that way, so the two helpers can disagree on the same input.Possible fix
+ const hasMiniToken = /(?:^|[-_/ ])mini(?:$|[-_/ ])/.test(normalizedName); + const hasNanoToken = /(?:^|[-_/ ])nano(?:$|[-_/ ])/.test(normalizedName);const isGpt54Mini = canonicalModelName === "gpt-5.4-mini" || - normalizedName.includes("gpt-5.4-mini") || - normalizedName.includes("gpt 5.4 mini"); + /\bgpt(?:-| )5\.4(?:-| )mini(?:\b|[- ])/.test(normalizedName); const isLightweight = !isGpt54Mini && !isCodexMini && - (normalizedName.includes("nano") || - normalizedName.includes("mini")); + (hasNanoToken || hasMiniToken);🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@lib/request/request-transformer.ts` around lines 480 - 514, The substring-based "mini" checks (used in isGpt54Mini, isCodexMini, isLightweight, etc.) incorrectly match suffixes like "-minimal"; change them to token-aware checks: compute tokens from normalizedName (e.g., split on non-alphanumeric boundaries) and test for the exact token "mini" (and "codex", "codex-mini" variants as separate tokens) rather than using normalizedName.includes("mini"). Update the definitions of isGpt54Mini, isCodexMini, isCodex, and isLightweight to use the token set (and preserve canonicalModelName checks like canonicalModelName === "gpt-5.4-mini") so "minimal" does not count as "mini" and the helpers agree with normalizeModel.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Outside diff comments:
In `@lib/request/request-transformer.ts`:
- Around line 480-514: The substring-based "mini" checks (used in isGpt54Mini,
isCodexMini, isLightweight, etc.) incorrectly match suffixes like "-minimal";
change them to token-aware checks: compute tokens from normalizedName (e.g.,
split on non-alphanumeric boundaries) and test for the exact token "mini" (and
"codex", "codex-mini" variants as separate tokens) rather than using
normalizedName.includes("mini"). Update the definitions of isGpt54Mini,
isCodexMini, isCodex, and isLightweight to use the token set (and preserve
canonicalModelName checks like canonicalModelName === "gpt-5.4-mini") so
"minimal" does not count as "mini" and the helpers agree with normalizeModel.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: cdd609ba-dd8a-4591-b4fa-a03daa3e160f
📒 Files selected for processing (1)
lib/request/request-transformer.ts
Contributor
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
lib/request/request-transformer.ts (1)
457-458:⚠️ Potential issue | 🟠 MajorProvider-prefixed aliases can be misclassified in reasoning support.
At Line 457,
normalizedNamekeeps the provider prefix, but Line 514 relies onstartsWith("gpt-5-").
This makesopenai/gpt-5failisGpt51General, which then affectssupportsNoneat Lines 545-549 and can incorrectly downgrade"none".💡 Proposed fix
-export function getReasoningConfig( +export function getReasoningConfig( modelName: string | undefined, userConfig: ConfigOptions = {}, ): ReasoningConfig { - const normalizedName = modelName?.toLowerCase() ?? ""; + const rawModelName = modelName?.toLowerCase() ?? ""; + const normalizedName = rawModelName.includes("/") + ? (rawModelName.split("/").pop() ?? rawModelName) + : rawModelName; const canonicalModelName = normalizeModel(modelName);Also applies to: 509-515, 545-549
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@lib/request/request-transformer.ts` around lines 457 - 458, normalizedName currently preserves provider prefixes (e.g., "openai/gpt-5") causing checks like isGpt51General that use startsWith("gpt-5-") to misclassify models; update the logic to compute normalizedName from the provider-agnostic canonical model (use normalizeModel(modelName) or canonicalModelName) or explicitly strip provider prefixes before performing startsWith checks so functions like isGpt51General and the supportsNone branch use the provider-agnostic name (refer to normalizedName, canonicalModelName, normalizeModel, and isGpt51General) to correctly detect gpt-5 variants and avoid downgrading "none".
🧹 Nitpick comments (1)
test/request-transformer.test.ts (1)
388-398: Add transform-level coverage for legacy aliasnonebehavior with provider prefixes.Great unit coverage for
getReasoningConfig, but this specific regression risk is in the full transform path (lookupModel+ body/provider layering).
Please add directtransformRequestBodyassertions forgpt-5-mini/openai/gpt-5-miniwithreasoning.effort: "none"to lock the downgrade behavior.✅ Suggested test addition
+it('should downgrade none to low for provider-prefixed legacy mini aliases in transformRequestBody', async () => { + const body: RequestBody = { + model: 'openai/gpt-5-mini', + input: [], + reasoning: { effort: 'none', summary: 'auto' }, + }; + const result = await transformRequestBody(body, codexInstructions); + expect(result.model).toBe('gpt-5.4'); + expect(result.reasoning?.effort).toBe('low'); +});Also applies to: 1695-1712
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@test/request-transformer.test.ts` around lines 388 - 398, Add transform-level tests that assert transformRequestBody preserves the legacy downgrade of reasoning.effort "none" to "low" when model aliases are used; specifically call transformRequestBody (or the full transform path that uses lookupModel and getReasoningConfig) with model set to "gpt-5-mini" and "openai/gpt-5-mini" and a body containing reasoning.effort: "none", then assert the resulting request body (or assembled reasoning config) has effort === "low". Ensure the test exercises the provider layering/lookupModel path rather than calling getReasoningConfig directly to lock the regression.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Outside diff comments:
In `@lib/request/request-transformer.ts`:
- Around line 457-458: normalizedName currently preserves provider prefixes
(e.g., "openai/gpt-5") causing checks like isGpt51General that use
startsWith("gpt-5-") to misclassify models; update the logic to compute
normalizedName from the provider-agnostic canonical model (use
normalizeModel(modelName) or canonicalModelName) or explicitly strip provider
prefixes before performing startsWith checks so functions like isGpt51General
and the supportsNone branch use the provider-agnostic name (refer to
normalizedName, canonicalModelName, normalizeModel, and isGpt51General) to
correctly detect gpt-5 variants and avoid downgrading "none".
---
Nitpick comments:
In `@test/request-transformer.test.ts`:
- Around line 388-398: Add transform-level tests that assert
transformRequestBody preserves the legacy downgrade of reasoning.effort "none"
to "low" when model aliases are used; specifically call transformRequestBody (or
the full transform path that uses lookupModel and getReasoningConfig) with model
set to "gpt-5-mini" and "openai/gpt-5-mini" and a body containing
reasoning.effort: "none", then assert the resulting request body (or assembled
reasoning config) has effort === "low". Ensure the test exercises the provider
layering/lookupModel path rather than calling getReasoningConfig directly to
lock the regression.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 64c8c197-c96c-44a6-a7ff-b96ee3d185b9
📒 Files selected for processing (3)
docs/configuration.mdlib/request/request-transformer.tstest/request-transformer.test.ts
🚧 Files skipped from review as they are similar to previous changes (1)
- docs/configuration.md
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
gpt-5.4-minias a first-class model family across normalization, prompt-family routing, config templates, and docsgpt-5-ministill maps togpt-5.4without inheriting first-class mini semanticsVerification
npm run typechecknpm run buildnpm testSummary by CodeRabbit
New Features
Documentation
Tests
note: greptile review for oc-chatgpt-multi-auth. cite files like
lib/foo.ts:123. confirm regression tests + windows concurrency/token redaction coverage.Greptile Summary
this pr promotes
gpt-5.4-minito a first-class model family across normalization, prompt routing, reasoning config, config templates, and docs — correctly isolated from both the basegpt-5.4family and the legacy lightweight aliases (gpt-5-mini,gpt-5-nano) that still map togpt-5.4without inheriting mini's semantics.key changes:
normalizeModelandgetModelFamilyboth correctly check for-minisuffix before falling through to the basegpt-5.4regex, preventing misclassificationgetReasoningConfigusescanonicalModelName === "gpt-5.4-mini"to gateisGpt54Mini, andsupportsNone/supportsXhigh/canonicalSupportsXhighare all extended correctlyisLightweightregex tightened to\bgpt(?:-| )5(?:-| )(?:mini|nano)(?:\b|[- ])/— correctly matchesgpt-5-mini/gpt-5-nanobut notgpt-5.4-mini(also guarded by!isGpt54Mini)prewarmCodexInstructionsaddsgpt-5.4-minito its default candidate list; isolated cache filegpt-5.4-mini-instructions.mdconfirmed by testhigh/xhighcorrectly setreasoningSummary: detailedgaps to address:
gpt-5-nanonone-blocking behavior is correct at runtime but has no vitest test — symmetric with thegpt-5-minitest added in this PRGPT_54_SNAPSHOT_DATEis shared for mini with no comment — document or split if mini uses a different snapshot cadenceConfidence Score: 4/5
Important Files Changed
Flowchart
%%{init: {'theme': 'neutral'}}%% flowchart TD A["raw model string\ne.g. 'openai/gpt-5.4-mini-high'"] --> B["normalizeModel()"] B --> C{"contains 'codex'?"} C -- yes --> D["→ gpt-5-codex"] C -- no --> E{"matches gpt-5.4-pro regex?"} E -- yes --> F["→ gpt-5.4-pro"] E -- no --> G{"matches gpt-5.4-mini regex?\n/gpt(?:-| )5\\.4(?:-| )mini/"} G -- yes --> H["→ gpt-5.4-mini ✨ NEW"] G -- no --> I{"matches gpt-5.4 regex?"} I -- yes --> J["→ gpt-5.4"] I -- no --> K["...other families..."] H --> L["getReasoningConfig()"] J --> L L --> M{"isGpt54Mini?\ncanonicalModelName === 'gpt-5.4-mini'"} M -- yes --> N["supportsNone = true\nsupportsXhigh = true\ndefaultEffort = 'high'"] M -- no --> O{"isLightweight?\n/gpt-5-mini|nano/ regex\n!isGpt54Mini guard"} O -- yes --> P["supportsNone = false\n→ 'none' upgraded to 'low'"] O -- no --> Q["normal gpt-5.4 path"]Prompt To Fix All With AI
Last reviewed commit: "fix(models): address..."
Context used: