Bump bundled @ai-sdk/openai-compatible 2.0.41 -> 2.0.75 (fixes fragmented reasoning streams) #50290

Activity

1. lscar commented on Sep 21, 2026

   

   lscar

   on Sep 21, 2026

   Author

   More actions

   Adding a self-contained repro that needs no captured traffic — only the trigger condition (a delta carrying reasoning_content together with an empty tool_calls: [] array, which is what DeepSeek/llama.cpp-style streams emit on every reasoning chunk).

   npm i ai@6 @ai-sdk/[email protected]   # rerun with @2.0.75

   repro.mjs:

   import { createOpenAICompatible } from "@ai-sdk/openai-compatible"
   import { streamText } from "ai"
   import { createRequire } from "node:module"
   
   const require = createRequire(import.meta.url)
   const pkgVersion = require("@ai-sdk/openai-compatible/package.json").version
   
   const TOKENS = ["We", " need", " to", " think", " first"]
   const sse = TOKENS.map(
     (t) =>
       `data: ${JSON.stringify({
         id: "chatcmpl-test",
         object: "chat.completion.chunk",
         choices: [
           {
             index: 0,
             delta: { role: "assistant", content: "", reasoning_content: t, tool_calls: [] },
             finish_reason: null,
           },
         ],
       })}`,
   ).concat([
     `data: ${JSON.stringify({
       id: "chatcmpl-test",
       object: "chat.completion.chunk",
       choices: [{ index: 0, delta: { content: "Because of Rayleigh scattering." }, finish_reason: "stop" }],
     })}`,
   ])
   
   const fetchMock = async () =>
     new Response(
       new ReadableStream({
         async start(ctl) {
           for (const line of sse) ctl.enqueue(new TextEncoder().encode(line + "\n\n"))
           ctl.enqueue(new TextEncoder().encode("data: [DONE]\n\n"))
           ctl.close()
         },
       }),
       { status: 200, headers: { "content-type": "text/event-stream" } },
     )
   
   const provider = createOpenAICompatible({ name: "replay", baseURL: "https://replay.local/v1", fetch: fetchMock })
   
   const counts = { "reasoning-start": 0, "reasoning-delta": 0, "reasoning-end": 0, "text-delta": 0, finish: 0 }
   const result = streamText({ model: provider.chatModel("test-model"), prompt: "hi" })
   for await (const part of result.fullStream) if (part.type in counts) counts[part.type]++
   
   console.log(`@ai-sdk/openai-compatible@${pkgVersion} + ai@${require("ai/package.json").version}`)
   console.log(`  reasoning tokens sent: ${TOKENS.length}`)
   console.log(`  ${JSON.stringify(counts)}`)

   Output — identical [email protected] core, only the provider differs:

   @ai-sdk/[email protected]:

   reasoning tokens sent: 5
   {"reasoning-start":5,"reasoning-delta":5,"reasoning-end":5,"text-delta":1,"finish":1}
   

   @ai-sdk/[email protected]:

   reasoning tokens sent: 5
   {"reasoning-start":1,"reasoning-delta":5,"reasoning-end":1,"text-delta":1,"finish":1}
   

   I also replayed a real captured DeepSeek stream (268 reasoning tokens, each delta carrying tool_calls: []):

   • 2.0.41 → reasoning-start:268 / reasoning-end:268
   • 2.0.75 → reasoning-start:1 / reasoning-end:1

   Both confirm the fix already present in 2.0.75 (f3c1980 — "keep reasoning streams contiguous when deltas include empty tool call arrays") is exactly what un-fragments these streams. Bumping the bundled version from 2.0.41 to 2.0.75 resolves it without touching the AI SDK 6 core.
2. thdxr commented on Oct 11, 2026

   

   thdxr

   on Oct 11, 2026

   Member

   More actions

   Thanks for the report! OpenCode 2.0 is now out and development is focused on V2, so we're closing issues filed against OpenCode 1. Please upgrade (opencode upgrade --major, details: https://anoma.ly/notes/opencode-2-0), and if this is still relevant on 2.0, reopen this issue or open a new one with your 2.0 version.

   — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊
3. thdxr

   closed this as not plannedon Oct 11, 2026