Skip to content

Commit 4e0509b

Browse files
Luis Daniel Riccio SilvaLuis Daniel Riccio Silva
Luis Daniel Riccio Silva
authored and
Luis Daniel Riccio Silva
committed
feat: add native OpenAI provider support for Codex Mini model
- Add codex-mini-latest model definition with proper pricing (.5/M input, /M output) - Implement support using existing v1/responses endpoint infrastructure (same as GPT-5) - Add comprehensive test coverage for streaming and non-streaming modes - Handle Codex Mini specific request format (instructions + input) - Reuse GPT-5's SSE parsing logic for efficient implementation Closes #5386
1 parent f53fd39 commit 4e0509b

File tree

3 files changed

+484
-10
lines changed

3 files changed

+484
-10
lines changed

‎packages/types/src/providers/openai.ts‎

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -220,6 +220,16 @@ export const openAiNativeModels = {
220220
outputPrice: 0.6,
221221
cacheReadsPrice: 0.075,
222222
},
223+
"codex-mini-latest": {
224+
maxTokens: 16_384,
225+
contextWindow: 200_000,
226+
supportsImages: false,
227+
supportsPromptCache: false,
228+
inputPrice: 1.5,
229+
outputPrice: 6,
230+
cacheReadsPrice: 0,
231+
description: "Codex Mini: Optimized coding model using v1/responses endpoint",
232+
},
223233
} as const satisfies Record<string, ModelInfo>
224234

225235
export const openAiModelInfoSaneDefaults: ModelInfo = {

‎src/api/providers/__tests__/openai-native.spec.ts‎

Lines changed: 264 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1514,4 +1514,268 @@ describe("GPT-5 streaming event coverage (additional)", () => {
15141514
// @ts-ignore
15151515
delete global.fetch
15161516
})
1517+
1518+
describe("Codex Mini Model", () => {
1519+
let handler: OpenAiNativeHandler
1520+
const mockOptions: ApiHandlerOptions = {
1521+
openAiNativeApiKey: "test-api-key",
1522+
apiModelId: "codex-mini-latest",
1523+
}
1524+
1525+
it("should handle codex-mini-latest streaming response", async () => {
1526+
// Mock fetch for Codex Mini responses API
1527+
const mockFetch = vitest.fn().mockResolvedValue({
1528+
ok: true,
1529+
body: new ReadableStream({
1530+
start(controller) {
1531+
// Codex Mini uses the same responses API format
1532+
controller.enqueue(
1533+
new TextEncoder().encode('data: {"type":"response.output_text.delta","delta":"Hello"}\n\n'),
1534+
)
1535+
controller.enqueue(
1536+
new TextEncoder().encode('data: {"type":"response.output_text.delta","delta":" from"}\n\n'),
1537+
)
1538+
controller.enqueue(
1539+
new TextEncoder().encode(
1540+
'data: {"type":"response.output_text.delta","delta":" Codex"}\n\n',
1541+
),
1542+
)
1543+
controller.enqueue(
1544+
new TextEncoder().encode(
1545+
'data: {"type":"response.output_text.delta","delta":" Mini!"}\n\n',
1546+
),
1547+
)
1548+
controller.enqueue(new TextEncoder().encode('data: {"type":"response.completed"}\n\n'))
1549+
controller.enqueue(new TextEncoder().encode("data: [DONE]\n\n"))
1550+
controller.close()
1551+
},
1552+
}),
1553+
})
1554+
global.fetch = mockFetch as any
1555+
1556+
handler = new OpenAiNativeHandler({
1557+
...mockOptions,
1558+
apiModelId: "codex-mini-latest",
1559+
})
1560+
1561+
const systemPrompt = "You are a helpful coding assistant."
1562+
const messages: Anthropic.Messages.MessageParam[] = [
1563+
{ role: "user", content: "Write a hello world function" },
1564+
]
1565+
1566+
const stream = handler.createMessage(systemPrompt, messages)
1567+
const chunks: any[] = []
1568+
for await (const chunk of stream) {
1569+
chunks.push(chunk)
1570+
}
1571+
1572+
// Verify text chunks
1573+
const textChunks = chunks.filter((c) => c.type === "text")
1574+
expect(textChunks).toHaveLength(4)
1575+
expect(textChunks.map((c) => c.text).join("")).toBe("Hello from Codex Mini!")
1576+
1577+
// Verify usage estimation (based on character count)
1578+
const usageChunks = chunks.filter((c) => c.type === "usage")
1579+
expect(usageChunks).toHaveLength(1)
1580+
expect(usageChunks[0]).toMatchObject({
1581+
type: "usage",
1582+
inputTokens: expect.any(Number),
1583+
outputTokens: expect.any(Number),
1584+
totalCost: expect.any(Number), // Codex Mini has pricing: $1.5/M input, $6/M output
1585+
})
1586+
1587+
// Verify cost is calculated correctly
1588+
expect(usageChunks[0].totalCost).toBeGreaterThan(0)
1589+
1590+
// Verify the request was made with correct parameters
1591+
expect(mockFetch).toHaveBeenCalledWith(
1592+
"https://api.openai.com/v1/responses",
1593+
expect.objectContaining({
1594+
method: "POST",
1595+
headers: expect.objectContaining({
1596+
"Content-Type": "application/json",
1597+
Authorization: "Bearer test-api-key",
1598+
Accept: "text/event-stream",
1599+
}),
1600+
body: expect.any(String),
1601+
}),
1602+
)
1603+
1604+
const requestBody = JSON.parse(mockFetch.mock.calls[0][1].body)
1605+
expect(requestBody).toMatchObject({
1606+
model: "codex-mini-latest",
1607+
instructions: systemPrompt,
1608+
input: "Write a hello world function",
1609+
stream: true,
1610+
})
1611+
1612+
// Clean up
1613+
delete (global as any).fetch
1614+
})
1615+
1616+
it("should handle codex-mini-latest non-streaming completion", async () => {
1617+
// Mock fetch for non-streaming response
1618+
const mockFetch = vitest.fn().mockResolvedValue({
1619+
ok: true,
1620+
json: async () => ({
1621+
output_text: "def hello_world():\n print('Hello, World!')",
1622+
}),
1623+
})
1624+
global.fetch = mockFetch as any
1625+
1626+
handler = new OpenAiNativeHandler({
1627+
...mockOptions,
1628+
apiModelId: "codex-mini-latest",
1629+
})
1630+
1631+
const result = await handler.completePrompt("Write a hello world function in Python")
1632+
1633+
expect(result).toBe("def hello_world():\n print('Hello, World!')")
1634+
1635+
// Verify the request
1636+
expect(mockFetch).toHaveBeenCalledWith(
1637+
"https://api.openai.com/v1/responses",
1638+
expect.objectContaining({
1639+
method: "POST",
1640+
headers: expect.objectContaining({
1641+
"Content-Type": "application/json",
1642+
Authorization: "Bearer test-api-key",
1643+
}),
1644+
body: expect.any(String),
1645+
}),
1646+
)
1647+
1648+
const requestBody = JSON.parse(mockFetch.mock.calls[0][1].body)
1649+
expect(requestBody).toMatchObject({
1650+
model: "codex-mini-latest",
1651+
instructions: "Complete the following prompt:",
1652+
input: "Write a hello world function in Python",
1653+
stream: false,
1654+
})
1655+
1656+
// Clean up
1657+
delete (global as any).fetch
1658+
})
1659+
1660+
it("should handle codex-mini-latest API errors", async () => {
1661+
// Mock fetch with error response
1662+
const mockFetch = vitest.fn().mockResolvedValue({
1663+
ok: false,
1664+
status: 429,
1665+
statusText: "Too Many Requests",
1666+
text: async () => "Rate limit exceeded",
1667+
})
1668+
global.fetch = mockFetch as any
1669+
1670+
handler = new OpenAiNativeHandler({
1671+
...mockOptions,
1672+
apiModelId: "codex-mini-latest",
1673+
})
1674+
1675+
const systemPrompt = "You are a helpful assistant."
1676+
const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Hello" }]
1677+
1678+
const stream = handler.createMessage(systemPrompt, messages)
1679+
1680+
// Should throw an error
1681+
await expect(async () => {
1682+
for await (const chunk of stream) {
1683+
// consume stream
1684+
}
1685+
}).rejects.toThrow("Codex Mini API request failed (429): Rate limit exceeded")
1686+
1687+
// Clean up
1688+
delete (global as any).fetch
1689+
})
1690+
1691+
it("should handle codex-mini-latest with multiple user messages", async () => {
1692+
// Mock fetch for streaming response
1693+
const mockFetch = vitest.fn().mockResolvedValue({
1694+
ok: true,
1695+
body: new ReadableStream({
1696+
start(controller) {
1697+
controller.enqueue(
1698+
new TextEncoder().encode(
1699+
'data: {"type":"response.output_text.delta","delta":"Combined response"}\n\n',
1700+
),
1701+
)
1702+
controller.enqueue(new TextEncoder().encode('data: {"type":"response.completed"}\n\n'))
1703+
controller.enqueue(new TextEncoder().encode("data: [DONE]\n\n"))
1704+
controller.close()
1705+
},
1706+
}),
1707+
})
1708+
global.fetch = mockFetch as any
1709+
1710+
handler = new OpenAiNativeHandler({
1711+
...mockOptions,
1712+
apiModelId: "codex-mini-latest",
1713+
})
1714+
1715+
const systemPrompt = "You are a helpful assistant."
1716+
const messages: Anthropic.Messages.MessageParam[] = [
1717+
{ role: "user", content: "First question" },
1718+
{ role: "assistant", content: "First answer" },
1719+
{ role: "user", content: "Second question" },
1720+
]
1721+
1722+
const stream = handler.createMessage(systemPrompt, messages)
1723+
const chunks: any[] = []
1724+
for await (const chunk of stream) {
1725+
chunks.push(chunk)
1726+
}
1727+
1728+
// Verify the request body only includes user messages
1729+
const requestBody = JSON.parse(mockFetch.mock.calls[0][1].body)
1730+
expect(requestBody.input).toBe("First question\n\nSecond question")
1731+
expect(requestBody.input).not.toContain("First answer")
1732+
1733+
// Clean up
1734+
delete (global as any).fetch
1735+
})
1736+
1737+
it("should handle codex-mini-latest stream error events", async () => {
1738+
// Mock fetch with error event in stream
1739+
const mockFetch = vitest.fn().mockResolvedValue({
1740+
ok: true,
1741+
body: new ReadableStream({
1742+
start(controller) {
1743+
controller.enqueue(
1744+
new TextEncoder().encode(
1745+
'data: {"type":"response.output_text.delta","delta":"Partial"}\n\n',
1746+
),
1747+
)
1748+
controller.enqueue(
1749+
new TextEncoder().encode(
1750+
'data: {"type":"response.error","error":{"message":"Model overloaded"}}\n\n',
1751+
),
1752+
)
1753+
controller.close()
1754+
},
1755+
}),
1756+
})
1757+
global.fetch = mockFetch as any
1758+
1759+
handler = new OpenAiNativeHandler({
1760+
...mockOptions,
1761+
apiModelId: "codex-mini-latest",
1762+
})
1763+
1764+
const systemPrompt = "You are a helpful assistant."
1765+
const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Hello" }]
1766+
1767+
const stream = handler.createMessage(systemPrompt, messages)
1768+
1769+
// Should throw an error when encountering error event
1770+
await expect(async () => {
1771+
const chunks = []
1772+
for await (const chunk of stream) {
1773+
chunks.push(chunk)
1774+
}
1775+
}).rejects.toThrow("Codex Mini stream error: Model overloaded")
1776+
1777+
// Clean up
1778+
delete (global as any).fetch
1779+
})
1780+
})
15171781
})

0 commit comments

Comments
 (0)