feat: add gpt-5.4.3 model + opencode compatibility updates

- add gpt-5.4/gpt-5.4-pro snapshot alias normalization and remap legacy gpt-5* aliases to gpt-5.4
- split gpt-5.4-pro into an isolated prompt family/cache key while preserving fallback behavior
- update OpenCode templates/docs for 1M context on gpt-5.4* and expand regression coverage across model mapping/family tests

Author: ndycode

CHANGELOG.md

@@ -25,6 +25,22 @@ all notable changes to this project. dates are ISO format (YYYY-MM-DD).
 - **doctor safe-fix edge path**: `codex-doctor fix` now reports a clear non-crashing message when no eligible account is available for auto-switch.
 - **first-time import flow**: `codex-import` no longer fails with `No accounts to export` when storage is empty; pre-import backup is skipped cleanly in zero-account setups.
 
+## [5.4.3] - 2026-03-06
+
+### added
+
+- **gpt-5.4 snapshot alias normalization**: added support for `gpt-5.4-2026-03-05*` and `gpt-5.4-pro-2026-03-05*` model IDs (including effort suffix variants).
+
+### changed
+
+- **legacy GPT-5 alias target updated**: `gpt-5`, `gpt-5-mini`, and `gpt-5-nano` now normalize to `gpt-5.4` as the default general family.
+- **gpt-5.4-pro family isolation**: prompt-family detection now keeps `gpt-5.4-pro` separate from `gpt-5.4` for independent prompt/cache handling while preserving fallback policy behavior (`gpt-5.4-pro -> gpt-5.4`).
+- **OpenCode 5.4 template limits updated**: shipped OpenCode config templates now set `gpt-5.4*` context to `1,000,000` (output remains `128,000`) and docs now include optional `model_context_window` / `model_auto_compact_token_limit` tuning guidance.
+
+### fixed
+
+- **5.4.3 regression/test coverage alignment**: expanded and corrected normalization, family-routing, and prompt-mapping tests for snapshot aliases, pro-family separation, and legacy alias behavior.
+
 ## [5.4.2] - 2026-03-05
 
 ### added

README.md

@@ -132,8 +132,8 @@ opencode run "Hello" --model=openai/gpt-5.4 --variant=medium
 
 | Model | Variants | Notes |
 |-------|----------|-------|
-| `gpt-5.4` | none, low, medium, high, xhigh | Latest GPT-5.4 with reasoning levels |
-| `gpt-5.4-pro` | low, medium, high, xhigh | Optional manual model for deeper reasoning; fallback default is `gpt-5.4-pro -> gpt-5.4` |
+| `gpt-5.4` | none, low, medium, high, xhigh | Latest GPT-5.4 with reasoning levels and 1,000,000 context window |
+| `gpt-5.4-pro` | low, medium, high, xhigh | Optional manual model for deeper reasoning; fallback default is `gpt-5.4-pro -> gpt-5.4` (also 1,000,000 context window) |
 | `gpt-5-codex` | low, medium, high | Canonical Codex model for code generation (default: high) |
 | `gpt-5.3-codex-spark` | low, medium, high, xhigh | Spark IDs are supported by the plugin, but access is entitlement-gated by account/workspace |
 | `gpt-5.1-codex-max` | low, medium, high, xhigh | Maximum context Codex |
@@ -143,6 +143,14 @@ opencode run "Hello" --model=openai/gpt-5.4 --variant=medium
 
 Config templates intentionally omit Spark model IDs by default to reduce entitlement failures on unsupported accounts. Add Spark manually only if your workspace is entitled.
 
+Legacy and snapshot aliases supported by the plugin:
+- `gpt-5`, `gpt-5-mini`, `gpt-5-nano` normalize to `gpt-5.4`
+- `gpt-5.4-2026-03-05` and `gpt-5.4-pro-2026-03-05` (including effort suffix variants) normalize to their stable families
+
+If your OpenCode runtime supports global auto-compaction settings, set:
+- `model_context_window = 1000000`
+- `model_auto_compact_token_limit = 900000`
+
 **Using variants:**
 ```bash
 # Modern OpenCode (v1.0.210+)
@@ -173,7 +181,7 @@ Add this to your `~/.config/opencode/opencode.json`:
       "models": {
         "gpt-5.4": {
           "name": "GPT 5.4 (OAuth)",
-          "limit": { "context": 272000, "output": 128000 },
+          "limit": { "context": 1000000, "output": 128000 },
           "modalities": { "input": ["text", "image", "pdf"], "output": ["text"] },
           "variants": {
             "none": { "reasoningEffort": "none" },

config/README.md

@@ -37,7 +37,11 @@ Both templates include:
 - GPT-5.4, GPT-5 Codex, GPT-5.1, GPT-5.1 Codex, GPT-5.1 Codex Max, GPT-5.1 Codex Mini
 - Reasoning variants per model family
 - `store: false` and `include: ["reasoning.encrypted_content"]`
-- Context metadata (272k context / 128k output)
+- Context metadata (`gpt-5.4*`: 1,000,000 context / 128,000 output; other shipped models: 272,000 / 128,000)
+
+If your OpenCode runtime supports global compaction tuning, you can also set:
+- `model_context_window = 1000000`
+- `model_auto_compact_token_limit = 900000`
 
 ## Spark model note
 

config/opencode-legacy.json

@@ -18,7 +18,7 @@
         "gpt-5.4-none": {
           "name": "GPT 5.4 None (OAuth)",
           "limit": {
-            "context": 272000,
+            "context": 1000000,
             "output": 128000
           },
           "modalities": {
@@ -43,7 +43,7 @@
         "gpt-5.4-low": {
           "name": "GPT 5.4 Low (OAuth)",
           "limit": {
-            "context": 272000,
+            "context": 1000000,
             "output": 128000
           },
           "modalities": {
@@ -68,7 +68,7 @@
         "gpt-5.4-medium": {
           "name": "GPT 5.4 Medium (OAuth)",
           "limit": {
-            "context": 272000,
+            "context": 1000000,
             "output": 128000
           },
           "modalities": {
@@ -93,7 +93,7 @@
         "gpt-5.4-high": {
           "name": "GPT 5.4 High (OAuth)",
           "limit": {
-            "context": 272000,
+            "context": 1000000,
             "output": 128000
           },
           "modalities": {
@@ -118,7 +118,7 @@
         "gpt-5.4-xhigh": {
           "name": "GPT 5.4 Extra High (OAuth)",
           "limit": {
-            "context": 272000,
+            "context": 1000000,
             "output": 128000
           },
           "modalities": {

config/opencode-modern.json

@@ -18,7 +18,7 @@
         "gpt-5.4": {
           "name": "GPT 5.4 (OAuth)",
           "limit": {
-            "context": 272000,
+            "context": 1000000,
             "output": 128000
           },
           "modalities": {

docs/configuration.md

@@ -46,6 +46,17 @@ controls how much thinking the model does.
 | `gpt-5.1` | none, low, medium, high |
 
 the shipped config templates include 21 presets and do not add optional IDs by default. add `gpt-5.4-pro` and/or `gpt-5.3-codex-spark` manually only for entitled workspaces.
+for context sizing, shipped templates use:
+- `gpt-5.4` and `gpt-5.4-pro`: `context=1000000`, `output=128000`
+- other shipped families: `context=272000`, `output=128000`
+
+model normalization aliases:
+- legacy `gpt-5`, `gpt-5-mini`, `gpt-5-nano` map to `gpt-5.4`
+- snapshot ids `gpt-5.4-2026-03-05*` and `gpt-5.4-pro-2026-03-05*` map to stable `gpt-5.4` / `gpt-5.4-pro`
+
+if your OpenCode runtime supports global compaction tuning, you can set:
+- `model_context_window = 1000000`
+- `model_auto_compact_token_limit = 900000`
 
 what they mean:
 - `none` - no reasoning phase (base models only; auto-converts to `low` for codex/pro families, including `gpt-5-codex` and `gpt-5.4-pro`)

docs/development/TESTING.md

@@ -54,10 +54,10 @@ npm test
 
 | User Selects | Plugin Receives | Normalizes To | Config Lookup | API Receives | Result |
 |--------------|-----------------|---------------|---------------|--------------|--------|
-| `openai/gpt-5` | `"gpt-5"` | `"gpt-5.1"` | `models["gpt-5"]` → undefined | `"gpt-5.1"` | ✅ Uses global options |
+| `openai/gpt-5` | `"gpt-5"` | `"gpt-5.4"` | `models["gpt-5"]` → undefined | `"gpt-5.4"` | ✅ Uses global options |
 | `openai/gpt-5-codex` | `"gpt-5-codex"` | `"gpt-5-codex"` | `models["gpt-5-codex"]` → undefined | `"gpt-5-codex"` | ✅ Uses global options |
-| `openai/gpt-5-mini` | `"gpt-5-mini"` | `"gpt-5.1"` | `models["gpt-5-mini"]` → undefined | `"gpt-5.1"` | ✅ Uses global options |
-| `openai/gpt-5-nano` | `"gpt-5-nano"` | `"gpt-5.1"` | `models["gpt-5-nano"]` → undefined | `"gpt-5.1"` | ✅ Uses global options |
+| `openai/gpt-5-mini` | `"gpt-5-mini"` | `"gpt-5.4"` | `models["gpt-5-mini"]` → undefined | `"gpt-5.4"` | ✅ Uses global options |
+| `openai/gpt-5-nano` | `"gpt-5-nano"` | `"gpt-5.4"` | `models["gpt-5-nano"]` → undefined | `"gpt-5.4"` | ✅ Uses global options |
 
 **Expected Behavior:**
 - ✅ All models work with global options
@@ -274,9 +274,9 @@ API receives: "gpt-5.1" ✅
 ```
 User selects: openai/my-gpt-5-variant
 Plugin receives: "my-gpt-5-variant"
-normalizeModel: "my-gpt-5-variant" → "gpt-5.1" ✅ (includes "gpt-5", not "codex")
+normalizeModel: "my-gpt-5-variant" → "gpt-5.4" ✅ (includes "gpt-5", not "codex")
 Config lookup: models["my-gpt-5-variant"] → Found ✅
-API receives: "gpt-5.1" ✅
+API receives: "gpt-5.4" ✅
 ```
 
 **Result:** ✅ Works (correct model selected)
@@ -617,9 +617,9 @@ normalizeModel("codex-mini-latest")     // → "gpt-5.1-codex-mini" ✅
 normalizeModel("gpt-5.1-codex")         // → "gpt-5-codex" ✅
 normalizeModel("gpt-5.1")               // → "gpt-5.1" ✅
 normalizeModel("my-codex-model")        // → "gpt-5-codex" ✅
-normalizeModel("gpt-5")                 // → "gpt-5.1" ✅
-normalizeModel("gpt-5-mini")            // → "gpt-5.1" ✅
-normalizeModel("gpt-5-nano")            // → "gpt-5.1" ✅
+normalizeModel("gpt-5")                 // → "gpt-5.4" ✅
+normalizeModel("gpt-5-mini")            // → "gpt-5.4" ✅
+normalizeModel("gpt-5-nano")            // → "gpt-5.4" ✅
 normalizeModel("GPT 5.4 Pro High")      // → "gpt-5.4-pro" ✅
 normalizeModel(undefined)                 // → "gpt-5.1" ✅
 normalizeModel("random-model")          // → "gpt-5.1" ✅ (fallback)
@@ -658,7 +658,7 @@ export function normalizeModel(model: string | undefined): string {
     return "gpt-5-codex";
   }
   if (normalized.includes("gpt-5") || normalized.includes("gpt 5")) {
-    return "gpt-5.1";
+    return "gpt-5.4";
   }
   return "gpt-5.1";
 }
@@ -667,7 +667,7 @@ export function normalizeModel(model: string | undefined): string {
 **Why this works:**
 - ✅ Case-insensitive (`.toLowerCase()` + `.includes()`)
 - ✅ Pattern-based (works with any naming)
-- ✅ Safe fallback (unknown models → `gpt-5.1`)
+- ✅ Legacy GPT-5 fallback (`gpt-5*` aliases → `gpt-5.4`) + safe unknown fallback (`gpt-5.1`)
 - ✅ Codex priority with explicit Codex Mini support (`codex-mini*` → `codex-mini-latest`)
 
 ---
@@ -723,17 +723,17 @@ opencode run "test" --model=openai/gpt-5-codex
 ```typescript
 describe('normalizeModel', () => {
   test('handles all default models', () => {
-    expect(normalizeModel('gpt-5')).toBe('gpt-5.1')
+    expect(normalizeModel('gpt-5')).toBe('gpt-5.4')
     expect(normalizeModel('gpt-5-codex')).toBe('gpt-5-codex')
     expect(normalizeModel('gpt-5-codex-mini')).toBe('gpt-5.1-codex-mini')
-    expect(normalizeModel('gpt-5-mini')).toBe('gpt-5.1')
-    expect(normalizeModel('gpt-5-nano')).toBe('gpt-5.1')
+    expect(normalizeModel('gpt-5-mini')).toBe('gpt-5.4')
+    expect(normalizeModel('gpt-5-nano')).toBe('gpt-5.4')
   })
 
   test('handles custom preset names', () => {
     expect(normalizeModel('gpt-5-codex-low')).toBe('gpt-5-codex')
     expect(normalizeModel('openai/gpt-5-codex-mini-high')).toBe('gpt-5.1-codex-mini')
-    expect(normalizeModel('gpt-5-high')).toBe('gpt-5.1')
+    expect(normalizeModel('gpt-5-high')).toBe('gpt-5.4')
   })
 
   test('handles legacy names', () => {

docs/getting-started.md

@@ -206,16 +206,16 @@ Safe mode effects:
 
 | Model | Variants | Notes |
 |-------|----------|-------|
-| `gpt-5.4` | none, low, medium, high, xhigh | Latest GPT-5.4 |
-| `gpt-5.4-pro` | low, medium, high, xhigh | Optional manual model for deeper reasoning; fallback default is `gpt-5.4-pro -> gpt-5.4` |
+| `gpt-5.4` | none, low, medium, high, xhigh | Latest GPT-5.4 (1,000,000 context) |
+| `gpt-5.4-pro` | low, medium, high, xhigh | Optional manual model for deeper reasoning; fallback default is `gpt-5.4-pro -> gpt-5.4` (1,000,000 context) |
 | `gpt-5-codex` | low, medium, high | Canonical Codex for code generation |
 | `gpt-5.3-codex-spark` | low, medium, high, xhigh | Optional manual model; entitlement-gated by account/workspace |
 | `gpt-5.1-codex-max` | low, medium, high, xhigh | Maximum context |
 | `gpt-5.1-codex` | low, medium, high | Standard Codex |
 | `gpt-5.1-codex-mini` | medium, high | Lightweight |
 | `gpt-5.1` | none, low, medium, high | Base model |
 
-**Total: 21 template presets** with 272k context / 128k output (+ optional Spark IDs when entitled).
+**Total: 21 template presets** with mixed context sizing: `gpt-5.4*` at 1,000,000 / 128,000 and other shipped families at 272,000 / 128,000 (+ optional Spark IDs when entitled).
 
 ---
 

lib/prompts/codex.ts

@@ -49,6 +49,7 @@ export type ModelFamily =
 	| "codex-max"
 	| "codex"
 	| "gpt-5.4"
+	| "gpt-5.4-pro"
 	| "gpt-5.2"
 	| "gpt-5.1";
 
@@ -61,6 +62,7 @@ export const MODEL_FAMILIES: readonly ModelFamily[] = [
 	"codex-max",
 	"codex",
 	"gpt-5.4",
+	"gpt-5.4-pro",
 	"gpt-5.2",
 	"gpt-5.1",
 ] as const;
@@ -75,6 +77,8 @@ const PROMPT_FILES: Record<ModelFamily, string> = {
 	codex: "gpt_5_codex_prompt.md",
 	// As of Codex rust-v0.111.0, GPT-5.4 uses the same prompt file family as GPT-5.2.
 	"gpt-5.4": "gpt_5_2_prompt.md",
+	// GPT-5.4-pro uses the same core prompt file as GPT-5.4, but keeps isolated cache/family state.
+	"gpt-5.4-pro": "gpt_5_2_prompt.md",
 	"gpt-5.2": "gpt_5_2_prompt.md",
 	"gpt-5.1": "gpt_5_1_prompt.md",
 };
@@ -87,6 +91,7 @@ const CACHE_FILES: Record<ModelFamily, string> = {
 	"codex-max": "codex-max-instructions.md",
 	codex: "codex-instructions.md",
 	"gpt-5.4": "gpt-5.4-instructions.md",
+	"gpt-5.4-pro": "gpt-5.4-pro-instructions.md",
 	"gpt-5.2": "gpt-5.2-instructions.md",
 	"gpt-5.1": "gpt-5.1-instructions.md",
 };
@@ -120,6 +125,9 @@ export function getModelFamily(normalizedModel: string): ModelFamily {
 	) {
 		return "codex";
 	}
+	if (/\bgpt(?:-| )5\.4(?:-| )pro(?:\b|[- ])/i.test(normalizedModel)) {
+		return "gpt-5.4-pro";
+	}
 	if (/\bgpt(?:-| )5\.4(?:\b|[- ])/i.test(normalizedModel)) {
 		return "gpt-5.4";
 	}
@@ -404,7 +412,7 @@ function refreshInstructionsInBackground(
  * Prewarm instruction caches for the provided models/families.
  */
 export function prewarmCodexInstructions(models: string[] = []): void {
-	const candidates = models.length > 0 ? models : ["gpt-5-codex", "gpt-5.4", "gpt-5.2", "gpt-5.1"];
+	const candidates = models.length > 0 ? models : ["gpt-5-codex", "gpt-5.4", "gpt-5.4-pro", "gpt-5.2", "gpt-5.1"];
 	for (const model of candidates) {
 		void getCodexInstructions(model).catch((error) => {
 			logDebug("Codex instruction prewarm failed", {

lib/request/helpers/model-map.ts

@@ -67,6 +67,12 @@ export const MODEL_MAP: Record<string, string> = {
 	"gpt-5.4-medium": "gpt-5.4",
 	"gpt-5.4-high": "gpt-5.4",
 	"gpt-5.4-xhigh": "gpt-5.4",
+	"gpt-5.4-2026-03-05": "gpt-5.4",
+	"gpt-5.4-2026-03-05-none": "gpt-5.4",
+	"gpt-5.4-2026-03-05-low": "gpt-5.4",
+	"gpt-5.4-2026-03-05-medium": "gpt-5.4",
+	"gpt-5.4-2026-03-05-high": "gpt-5.4",
+	"gpt-5.4-2026-03-05-xhigh": "gpt-5.4",
 
 	// ============================================================================
 	// GPT-5.4 Pro Models (optional/manual config)
@@ -77,6 +83,12 @@ export const MODEL_MAP: Record<string, string> = {
 	"gpt-5.4-pro-medium": "gpt-5.4-pro",
 	"gpt-5.4-pro-high": "gpt-5.4-pro",
 	"gpt-5.4-pro-xhigh": "gpt-5.4-pro",
+	"gpt-5.4-pro-2026-03-05": "gpt-5.4-pro",
+	"gpt-5.4-pro-2026-03-05-none": "gpt-5.4-pro",
+	"gpt-5.4-pro-2026-03-05-low": "gpt-5.4-pro",
+	"gpt-5.4-pro-2026-03-05-medium": "gpt-5.4-pro",
+	"gpt-5.4-pro-2026-03-05-high": "gpt-5.4-pro",
+	"gpt-5.4-pro-2026-03-05-xhigh": "gpt-5.4-pro",
 
 	// ============================================================================
 	// GPT-5.2 Models (supports none/low/medium/high/xhigh per OpenAI API docs)
@@ -128,11 +140,11 @@ export const MODEL_MAP: Record<string, string> = {
 	"gpt-5-codex-mini-high": "gpt-5.1-codex-mini",
 
 	// ============================================================================
-	// GPT-5 General Purpose Models (LEGACY - maps to gpt-5.1 as gpt-5 is being phased out)
+	// GPT-5 General Purpose Models (LEGACY - maps to gpt-5.4 latest)
 	// ============================================================================
-	"gpt-5": "gpt-5.1",
-	"gpt-5-mini": "gpt-5.1",
-	"gpt-5-nano": "gpt-5.1",
+	"gpt-5": "gpt-5.4",
+	"gpt-5-mini": "gpt-5.4",
+	"gpt-5-nano": "gpt-5.4",
 };
 
 /**