Add Codex Luna and preview routing

commit: 9e3f5d24390f4a882ffeafd2844627ecb3d41949
author: patriceckhart · 3mo ago · 2026-07-09 18:42 UTC
parent: c6b35a2
signature: unsigned
refs: tag: v0.2.70

4 files · +139 -16

M packages/provider/models.go +6 -12


@@ -316,18 +316,6 @@ var Catalog = []Model{
316	316		PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
317	317		Speculative: true,
318	318		},
319		-	{
320		-	Provider: "openai", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol",
321		-	ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true,
322		-	PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
323		-	Speculative: true,
324		-	},
325		-	{
326		-	Provider: "openai", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra",
327		-	ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true,
328		-	PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
329		-	Speculative: true,
330		-	},
331	319
332	320		// ---- OpenAI Codex / ChatGPT subscription backend ----
333	321		// Same model ids as the OpenAI family, but routed through the

@@ -356,6 +344,12 @@ var Catalog = []Model{
356	344		PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
357	345		Speculative: true,
358	346		},
	347	+	{
	348	+	Provider: "openai-codex", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna",
	349	+	ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true,
	350	+	PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1,
	351	+	Speculative: true,
	352	+	},
359	353		{
360	354		Provider: "openai-codex", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol",
361	355		ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true,

 		PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
 		Speculative: true,
 	},
-	{
-		Provider: "openai", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol",
-		ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true,
-		PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
-		Speculative: true,
-	},
-	{
-		Provider: "openai", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra",
-		ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true,
-		PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
-		Speculative: true,
-	},
 	// ---- OpenAI Codex / ChatGPT subscription backend ----
 	// Same model ids as the OpenAI family, but routed through the
 		PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
 		Speculative: true,
 	},
+	{
+		Provider: "openai-codex", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna",
+		ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true,
+		PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1,
+		Speculative: true,
+	},
 	{
 		Provider: "openai-codex", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol",
 		ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true,

M packages/provider/openai_codex.go +42 -2


@@ -3,7 +3,9 @@ package provider
3	3		import (
4	4		"bytes"
5	5		"context"
	6	+	"crypto/rand"
6	7		"encoding/base64"
	8	+	"encoding/hex"
7	9		"encoding/json"
8	10		"fmt"
9	11		"io"

@@ -146,6 +148,7 @@ type codexRequest struct {
146	148		ParallelToolCalls bool `json:"parallel_tool_calls"`
147	149		Include []string `json:"include,omitempty"`
148	150		Reasoning *codexReasoningConfig `json:"reasoning,omitempty"`
	151	+	PromptCacheKey string `json:"prompt_cache_key,omitempty"`
149	152		}
150	153
151	154		// ---- Request building ----

@@ -310,6 +313,29 @@ func splitCallID(id string) (string, string) {
310	313		return id, ""
311	314		}
312	315
	316	+	func usesCodexCLIRouting(model string) bool {
	317	+	switch model {
	318	+	case "gpt-5.6-luna", "gpt-5.6-luna-pro", "gpt-5.6-terra", "gpt-5.6-terra-pro":
	319	+	return true
	320	+	default:
	321	+	return false
	322	+	}
	323	+	}
	324	+
	325	+	func newCodexSessionID() string {
	326	+	var b [16]byte
	327	+	if _, err := rand.Read(b[:]); err != nil {
	328	+	return fmt.Sprintf("zot-%d", time.Now().UnixNano())
	329	+	}
	330	+	return strings.Join([]string{
	331	+	hex.EncodeToString(b[0:4]),
	332	+	hex.EncodeToString(b[4:6]),
	333	+	hex.EncodeToString(b[6:8]),
	334	+	hex.EncodeToString(b[8:10]),
	335	+	hex.EncodeToString(b[10:16]),
	336	+	}, "-")
	337	+	}
	338	+
313	339		// ---- Streaming ----
314	340
315	341		func (c *codexClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {

@@ -317,6 +343,11 @@ func (c *codexClient) Stream(ctx context.Context, req Request) (<-chan Event, er
317	343		if err != nil {
318	344		return nil, err
319	345		}
	346	+	var codexCLISessionID string
	347	+	if usesCodexCLIRouting(wire.Model) {
	348	+	codexCLISessionID = newCodexSessionID()
	349	+	wire.PromptCacheKey = codexCLISessionID
	350	+	}
320	351		body, err := json.Marshal(wire)
321	352		if err != nil {
322	353		return nil, err

@@ -332,8 +363,17 @@ func (c *codexClient) Stream(ctx context.Context, req Request) (<-chan Event, er
332	363		httpReq.Header.Set("authorization", "Bearer "+c.token)
333	364		httpReq.Header.Set("chatgpt-account-id", c.accountID)
334	365		httpReq.Header.Set("openai-beta", "responses=experimental")
335		-	httpReq.Header.Set("originator", "zot")
336		-	httpReq.Header.Set("user-agent", fmt.Sprintf("zot (%s %s)", runtime.GOOS, runtime.GOARCH))
	366	+	if codexCLISessionID != "" {
	367	+	// Some preview models are only admitted or reliably served by the
	368	+	// ChatGPT Codex backend when the request follows Codex CLI routing
	369	+	// metadata. Keep this narrow so Sol retains zot's proven shape.
	370	+	httpReq.Header.Set("originator", "codex_cli_rs")
	371	+	httpReq.Header.Set("session-id", codexCLISessionID)
	372	+	httpReq.Header.Set("user-agent", "codex_cli_rs/0.0.0")
	373	+	} else {
	374	+	httpReq.Header.Set("originator", "zot")
	375	+	httpReq.Header.Set("user-agent", fmt.Sprintf("zot (%s %s)", runtime.GOOS, runtime.GOARCH))
	376	+	}
337	377		return httpReq, nil
338	378		}
339	379


@@ -3,7 +3,9 @@ package provider
3		import (	3		import (
4		"bytes"	4		"bytes"
5		"context"	5		"context"
			6	+	"crypto/rand"
6		"encoding/base64"	7		"encoding/base64"
			8	+	"encoding/hex"
7		"encoding/json"	9		"encoding/json"
8		"fmt"	10		"fmt"
9		"io"	11		"io"

@@ -146,6 +148,7 @@ type codexRequest struct {
146		ParallelToolCalls bool `json:"parallel_tool_calls"`	148		ParallelToolCalls bool `json:"parallel_tool_calls"`
147		Include []string `json:"include,omitempty"`	149		Include []string `json:"include,omitempty"`
148		Reasoning *codexReasoningConfig `json:"reasoning,omitempty"`	150		Reasoning *codexReasoningConfig `json:"reasoning,omitempty"`
			151	+	PromptCacheKey string `json:"prompt_cache_key,omitempty"`
149		}	152		}
150			153
151		// ---- Request building ----	154		// ---- Request building ----

@@ -310,6 +313,29 @@ func splitCallID(id string) (string, string) {
310		return id, ""	313		return id, ""
311		}	314		}
312			315
			316	+	func usesCodexCLIRouting(model string) bool {
			317	+	switch model {
			318	+	case "gpt-5.6-luna", "gpt-5.6-luna-pro", "gpt-5.6-terra", "gpt-5.6-terra-pro":
			319	+	return true
			320	+	default:
			321	+	return false
			322	+	}
			323	+	}
			324	+
			325	+	func newCodexSessionID() string {
			326	+	var b [16]byte
			327	+	if _, err := rand.Read(b[:]); err != nil {
			328	+	return fmt.Sprintf("zot-%d", time.Now().UnixNano())
			329	+	}
			330	+	return strings.Join([]string{
			331	+	hex.EncodeToString(b[0:4]),
			332	+	hex.EncodeToString(b[4:6]),
			333	+	hex.EncodeToString(b[6:8]),
			334	+	hex.EncodeToString(b[8:10]),
			335	+	hex.EncodeToString(b[10:16]),
			336	+	}, "-")
			337	+	}
			338	+
313		// ---- Streaming ----	339		// ---- Streaming ----
314			340
315		func (c *codexClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {	341		func (c *codexClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {

@@ -317,6 +343,11 @@ func (c *codexClient) Stream(ctx context.Context, req Request) (<-chan Event, er
317		if err != nil {	343		if err != nil {
318		return nil, err	344		return nil, err
319		}	345		}
			346	+	var codexCLISessionID string
			347	+	if usesCodexCLIRouting(wire.Model) {
			348	+	codexCLISessionID = newCodexSessionID()
			349	+	wire.PromptCacheKey = codexCLISessionID
			350	+	}
320		body, err := json.Marshal(wire)	351		body, err := json.Marshal(wire)
321		if err != nil {	352		if err != nil {
322		return nil, err	353		return nil, err

@@ -332,8 +363,17 @@ func (c *codexClient) Stream(ctx context.Context, req Request) (<-chan Event, er
332		httpReq.Header.Set("authorization", "Bearer "+c.token)	363		httpReq.Header.Set("authorization", "Bearer "+c.token)
333		httpReq.Header.Set("chatgpt-account-id", c.accountID)	364		httpReq.Header.Set("chatgpt-account-id", c.accountID)
334		httpReq.Header.Set("openai-beta", "responses=experimental")	365		httpReq.Header.Set("openai-beta", "responses=experimental")
335	-	httpReq.Header.Set("originator", "zot")	366	+	if codexCLISessionID != "" {
336	-	httpReq.Header.Set("user-agent", fmt.Sprintf("zot (%s %s)", runtime.GOOS, runtime.GOARCH))	367	+	// Some preview models are only admitted or reliably served by the
			368	+	// ChatGPT Codex backend when the request follows Codex CLI routing
			369	+	// metadata. Keep this narrow so Sol retains zot's proven shape.
			370	+	httpReq.Header.Set("originator", "codex_cli_rs")
			371	+	httpReq.Header.Set("session-id", codexCLISessionID)
			372	+	httpReq.Header.Set("user-agent", "codex_cli_rs/0.0.0")
			373	+	} else {
			374	+	httpReq.Header.Set("originator", "zot")
			375	+	httpReq.Header.Set("user-agent", fmt.Sprintf("zot (%s %s)", runtime.GOOS, runtime.GOARCH))
			376	+	}
337		return httpReq, nil	377		return httpReq, nil
338		}	378		}
339			379

M packages/provider/openai_codex_test.go +90 -0

@@ -1,6 +1,11 @@
1	1		package provider
2	2
3	3		import (
	4	+	"bytes"
	5	+	"context"
	6	+	"encoding/json"
	7	+	"io"
	8	+	"net/http"
4	9		"strings"
5	10		"testing"
6	11		)

@@ -61,3 +66,88 @@ func TestCodexImageToolResultMirror(t *testing.T) {
61	66		t.Fatalf("mirrored user image was not serialized as input_image")
62	67		}
63	68		}
	69	+
	70	+	type roundTripFunc func(http.Request) (http.Response, error)
	71	+
	72	+	func (f roundTripFunc) RoundTrip(r http.Request) (http.Response, error) { return f(r) }
	73	+
	74	+	func TestCodexPreviewModelUsesCodexCLIShape(t *testing.T) {
	75	+	c := NewOpenAICodex("token", "acct", "https://example.test/backend-api/codex/responses").(*codexClient)
	76	+	var gotReq *http.Request
	77	+	var gotBody codexRequest
	78	+	c.http.Transport = roundTripFunc(func(r http.Request) (http.Response, error) {
	79	+	gotReq = r
	80	+	if err := json.NewDecoder(r.Body).Decode(&gotBody); err != nil {
	81	+	t.Fatal(err)
	82	+	}
	83	+	return &http.Response{
	84	+	StatusCode: http.StatusOK,
	85	+	Header: http.Header{"Content-Type": []string{"text/event-stream"}},
	86	+	Body: io.NopCloser(strings.NewReader("")),
	87	+	}, nil
	88	+	})
	89	+
	90	+	events, err := c.Stream(context.Background(), Request{
	91	+	Model: "gpt-5.6-terra",
	92	+	Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
	93	+	})
	94	+	if err != nil {
	95	+	t.Fatal(err)
	96	+	}
	97	+	for range events {
	98	+	}
	99	+
	100	+	if gotReq == nil {
	101	+	t.Fatal("request was not sent")
	102	+	}
	103	+	if gotReq.Header.Get("originator") != "codex_cli_rs" {
	104	+	t.Fatalf("originator = %q", gotReq.Header.Get("originator"))
	105	+	}
	106	+	if gotReq.Header.Get("user-agent") != "codex_cli_rs/0.0.0" {
	107	+	t.Fatalf("user-agent = %q", gotReq.Header.Get("user-agent"))
	108	+	}
	109	+	if gotBody.PromptCacheKey == "" {
	110	+	t.Fatal("prompt_cache_key was not set")
	111	+	}
	112	+	if gotReq.Header.Get("session-id") != gotBody.PromptCacheKey {
	113	+	t.Fatalf("session-id = %q, prompt_cache_key = %q", gotReq.Header.Get("session-id"), gotBody.PromptCacheKey)
	114	+	}
	115	+	}
	116	+
	117	+	func TestCodexSolKeepsZotShape(t *testing.T) {
	118	+	c := NewOpenAICodex("token", "acct", "https://example.test/backend-api/codex/responses").(*codexClient)
	119	+	var gotReq *http.Request
	120	+	var body bytes.Buffer
	121	+	c.http.Transport = roundTripFunc(func(r http.Request) (http.Response, error) {
	122	+	gotReq = r
	123	+	_, _ = body.ReadFrom(r.Body)
	124	+	return &http.Response{
	125	+	StatusCode: http.StatusOK,
	126	+	Header: http.Header{"Content-Type": []string{"text/event-stream"}},
	127	+	Body: io.NopCloser(strings.NewReader("")),
	128	+	}, nil
	129	+	})
	130	+
	131	+	events, err := c.Stream(context.Background(), Request{
	132	+	Model: "gpt-5.6-sol",
	133	+	Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
	134	+	})
	135	+	if err != nil {
	136	+	t.Fatal(err)
	137	+	}
	138	+	for range events {
	139	+	}
	140	+
	141	+	if gotReq == nil {
	142	+	t.Fatal("request was not sent")
	143	+	}
	144	+	if gotReq.Header.Get("originator") != "zot" {
	145	+	t.Fatalf("originator = %q", gotReq.Header.Get("originator"))
	146	+	}
	147	+	if gotReq.Header.Get("session-id") != "" {
	148	+	t.Fatalf("session-id = %q", gotReq.Header.Get("session-id"))
	149	+	}
	150	+	if strings.Contains(body.String(), "prompt_cache_key") {
	151	+	t.Fatalf("Sol request unexpectedly included prompt_cache_key: %s", body.String())
	152	+	}
	153	+	}

@@ -1,6 +1,11 @@
1	package provider	1		package provider
2		2
3	import (	3		import (
		4	+	"bytes"
		5	+	"context"
		6	+	"encoding/json"
		7	+	"io"
		8	+	"net/http"
4	"strings"	9		"strings"
5	"testing"	10		"testing"
6	)	11		)

@@ -61,3 +66,88 @@ func TestCodexImageToolResultMirror(t *testing.T) {
61	t.Fatalf("mirrored user image was not serialized as input_image")	66		t.Fatalf("mirrored user image was not serialized as input_image")
62	}	67		}
63	}	68		}
		69	+
		70	+	type roundTripFunc func(http.Request) (http.Response, error)
		71	+
		72	+	func (f roundTripFunc) RoundTrip(r http.Request) (http.Response, error) { return f(r) }
		73	+
		74	+	func TestCodexPreviewModelUsesCodexCLIShape(t *testing.T) {
		75	+	c := NewOpenAICodex("token", "acct", "https://example.test/backend-api/codex/responses").(*codexClient)
		76	+	var gotReq *http.Request
		77	+	var gotBody codexRequest
		78	+	c.http.Transport = roundTripFunc(func(r http.Request) (http.Response, error) {
		79	+	gotReq = r
		80	+	if err := json.NewDecoder(r.Body).Decode(&gotBody); err != nil {
		81	+	t.Fatal(err)
		82	+	}
		83	+	return &http.Response{
		84	+	StatusCode: http.StatusOK,
		85	+	Header: http.Header{"Content-Type": []string{"text/event-stream"}},
		86	+	Body: io.NopCloser(strings.NewReader("")),
		87	+	}, nil
		88	+	})
		89	+
		90	+	events, err := c.Stream(context.Background(), Request{
		91	+	Model: "gpt-5.6-terra",
		92	+	Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
		93	+	})
		94	+	if err != nil {
		95	+	t.Fatal(err)
		96	+	}
		97	+	for range events {
		98	+	}
		99	+
		100	+	if gotReq == nil {
		101	+	t.Fatal("request was not sent")
		102	+	}
		103	+	if gotReq.Header.Get("originator") != "codex_cli_rs" {
		104	+	t.Fatalf("originator = %q", gotReq.Header.Get("originator"))
		105	+	}
		106	+	if gotReq.Header.Get("user-agent") != "codex_cli_rs/0.0.0" {
		107	+	t.Fatalf("user-agent = %q", gotReq.Header.Get("user-agent"))
		108	+	}
		109	+	if gotBody.PromptCacheKey == "" {
		110	+	t.Fatal("prompt_cache_key was not set")
		111	+	}
		112	+	if gotReq.Header.Get("session-id") != gotBody.PromptCacheKey {
		113	+	t.Fatalf("session-id = %q, prompt_cache_key = %q", gotReq.Header.Get("session-id"), gotBody.PromptCacheKey)
		114	+	}
		115	+	}
		116	+
		117	+	func TestCodexSolKeepsZotShape(t *testing.T) {
		118	+	c := NewOpenAICodex("token", "acct", "https://example.test/backend-api/codex/responses").(*codexClient)
		119	+	var gotReq *http.Request
		120	+	var body bytes.Buffer
		121	+	c.http.Transport = roundTripFunc(func(r http.Request) (http.Response, error) {
		122	+	gotReq = r
		123	+	_, _ = body.ReadFrom(r.Body)
		124	+	return &http.Response{
		125	+	StatusCode: http.StatusOK,
		126	+	Header: http.Header{"Content-Type": []string{"text/event-stream"}},
		127	+	Body: io.NopCloser(strings.NewReader("")),
		128	+	}, nil
		129	+	})
		130	+
		131	+	events, err := c.Stream(context.Background(), Request{
		132	+	Model: "gpt-5.6-sol",
		133	+	Messages: []Message{{Role: RoleUser, Content: []Content{TextBlock{Text: "hi"}}}},
		134	+	})
		135	+	if err != nil {
		136	+	t.Fatal(err)
		137	+	}
		138	+	for range events {
		139	+	}
		140	+
		141	+	if gotReq == nil {
		142	+	t.Fatal("request was not sent")
		143	+	}
		144	+	if gotReq.Header.Get("originator") != "zot" {
		145	+	t.Fatalf("originator = %q", gotReq.Header.Get("originator"))
		146	+	}
		147	+	if gotReq.Header.Get("session-id") != "" {
		148	+	t.Fatalf("session-id = %q", gotReq.Header.Get("session-id"))
		149	+	}
		150	+	if strings.Contains(body.String(), "prompt_cache_key") {
		151	+	t.Fatalf("Sol request unexpectedly included prompt_cache_key: %s", body.String())
		152	+	}
		153	+	}

M packages/provider/provider_test.go +1 -2


@@ -393,8 +393,7 @@ func TestGPT56CatalogEntries(t *testing.T) {
393	393		priceOut float64
394	394		cacheRead float64
395	395		}{
396		-	{"openai", "gpt-5.6-sol", 5, 30, 0.5},
397		-	{"openai", "gpt-5.6-terra", 2.5, 15, 0.25},
	396	+	{"openai-codex", "gpt-5.6-luna", 1, 6, 0.1},
398	397		{"openai-codex", "gpt-5.6-sol", 5, 30, 0.5},
399	398		{"openai-codex", "gpt-5.6-terra", 2.5, 15, 0.25},
400	399		}

 		priceOut   float64
 		cacheRead  float64
 	}{
-		{"openai", "gpt-5.6-sol", 5, 30, 0.5},
+		{"openai-codex", "gpt-5.6-luna", 1, 6, 0.1},
-		{"openai", "gpt-5.6-terra", 2.5, 15, 0.25},
 		{"openai-codex", "gpt-5.6-sol", 5, 30, 0.5},
 		{"openai-codex", "gpt-5.6-terra", 2.5, 15, 0.25},
 	}