Repository navigation
[Provider compatibility] Qwen models fail through Responses → Chat translation: "System message must be at the beginning" #6242
Description
Client or integration
Codex CLI
Area
Provider adapter
Summary
Qwen models cannot be used through OpenCodex when routed from the Codex CLI.
The upstream provider returns HTTP 502 with:
upstream_server_error
System message must be at the beginning.
The OpenCodex log shows that the request is being processed through the translated protocol path:
Client API: Responses
Delivery mode: Translated
Upstream API: Chat
Request path: Responses → IR → Chat
Response path: Chat → IR → Responses
This appears to indicate that the Responses → IR → Chat translation is producing a Chat Completions message sequence where the system message is not the first message.
Importantly, the same Qwen model works when called directly against the provider without going through OpenCodex, so the model/provider itself appears functional.
Reproduction
- Configure OpenCodex with the AMD or any provider and with a Qwen model.
- Start OpenCodex and use the model through Codex CLI.
- Send a normal prompt.
- The request fails with HTTP 502.
Observed error:
System message must be at the beginning.
The OpenCodex log identifies the selected route as:
AMD/Qwen3.8-27B — explicit-provider-namespace
and the protocol path as:
Responses → IR → Chat
- Call the same Qwen model directly against the provider, bypassing OpenCodex.
- The direct request succeeds.
Version
2.71.0
Operating system
Kali GNU/Linux 2026.3
Provider and model
No response
Logs or error output
HTTP 502 Bad upstream response
Failure cause:
Upstream fault (reached: protocol prelude)
Upstream reason:
System message must be at the beginning.
Provider:
AMD
Model:
Qwen3.8-27B
Protocol:
Responses → IR → Chat
{
"requestId": "ocx-da488b5f94e4984c782d743f096f9993",
"timestamp": 1790691239442,
"model": "Qwen3.8-27B",
"provider": "AMD",
"admissionKind": "loopback",
"inboundProtocol": "responses",
"accountLogLabel": "kb697860e8b4c1dd0df9938d2160aeca1",
"conversationId": "d17ab8bcfdcd3a437fa74a3f441ea61b",
"requestedModel": "AMD/Qwen3.8-27B",
"requestedAlias": "AMD/Qwen3.8-27B",
"requestedEffort": "xhigh",
"effectiveEffort": "xhigh",
"reasoningWireField": "reasoning_effort",
"reasoningWireValue": "xhigh",
"configuredServiceTier": "fast",
"configuredSpeedLabel": "fast",
"modelSupportsServiceTier": false,
"tierOutcome": {
"wireKind": null,
"wireValue": null,
"fastOutcome": "unknown",
"confirmation": "unknown"
},
"resolvedModel": "Qwen3.8-27B",
"servedModel": "Qwen3.8-27B",
"status": 502,
"durationMs": 2385,
"errorCode": "upstream_server_error",
"terminalStatus": "failed",
"closeReason": "terminal",
"upstreamError": "System message must be at the beginning.",
"usageStatus": "unreported",
"attempts": [
{
"ordinal": 1,
"provider": "AMD",
"model": "Qwen3.8-27B",
"adapter": "openai-chat",
"status": 502,
"durationMs": 2372,
"sendCount": 1,
"recoveryKinds": [],
"usageStatus": "unreported",
"requestedEffort": "xhigh",
"effectiveEffort": "xhigh",
"reasoningWireField": "reasoning_effort",
"reasoningWireValue": "xhigh",
"tierOutcome": {
"wireKind": null,
"wireValue": null,
"fastOutcome": "unknown",
"confirmation": "unknown"
},
"accountLogLabel": "kb697860e8b4c1dd0df9938d2160aeca1",
"deliverySummary": {
"adapterEvents": 1,
"relayedEvents": 3,
"semanticBytes": 0,
"sideEffectEvents": 0,
"terminalEvents": 1
},
"errorCode": "upstream_server_error",
"failureStage": "protocol-prelude",
"failureCause": "upstream-fault",
"resendPermission": "refused-ambiguous",
"displayMetrics": {
"tokPerSecond": {
"kind": "unavailable",
"reason": "usage_missing"
},
"decodeTokPerSecond": {
"kind": "unavailable",
"reason": "usage_missing"
},
"cost": {
"kind": "unavailable",
"reason": "usage_missing"
}
}
}
],
"spend": {
"sends": 1,
"settled": 1,
"unresolved": 0
},
"transportPhase": "terminal_sse",
"terminalSource": "upstream",
"routeDecision": {
"version": 1,
"decisionId": "278cb2b49b0c",
"createdAt": 1790691239454,
"requestedModel": "AMD/Qwen3.8-27B",
"routeKind": "explicit-provider",
"requirements": [],
"candidates": [
{
"provider": "AMD",
"model": "Qwen3.8-27B",
"eligible": true,
"exclusions": []
}
],
"selected": {
"candidateIndex": 0,
"provider": "AMD",
"model": "Qwen3.8-27B",
"reason": "explicit-provider-namespace"
}
},
"protocolTrace": {
"v": 1,
"inbound": "responses",
"mode": "translated",
"upstream": "chat",
"requestPath": [
"responses",
"ir",
"chat"
],
"responsePath": [
"chat",
"ir",
"responses"
],
"reasonCodes": [
"cross-wire-ir"
],
"attempts": [
{
"ordinal": 1,
"upstream": "chat",
"mode": "translated",
"requestPath": [
"responses",
"ir",
"chat"
]
}
],
"contractVersion": "2026-09-25.2"
},
"failureStage": "protocol-prelude",
"failureCause": "upstream-fault",
"resendPermission": "refused-ambiguous",
"displayMetrics": {
"tokPerSecond": {
"kind": "unavailable",
"reason": "usage_missing"
},
"decodeTokPerSecond": {
"kind": "unavailable",
"reason": "usage_missing"
},
"cost": {
"kind": "unavailable",
"reason": "combo_attempt_unavailable"
}
}
}Screenshots and supporting files
Redacted configuration
Checks
- I searched existing issues and documentation.
- I removed secrets, tokens, account details, request credentials, and personal data.
Activity
lidge-jun commentedon Sep 29, 2026
Fixed on dev by #6249. The Qwen3.8 chat template rejects a system message after the first turn (
System message must be at the beginning.) and does not acceptdeveloper, so for translated Qwen3.8-27B routes a later instruction is now sent as a user message in its original position; other destinations are unchanged. Not verified against a live AMD endpoint. Ships in the next release; please reopen with a redacted request log if the error persists.Accept later system messages in the Qwen3.8 template
K9ineX3D commentedon Oct 1, 2026
This issue still persist for the model
Qwen3.8-Flash-Nextno matter the provider or harness. You can check it via any provider.Btw 3.8 27B works
Raw logs has been provided below.
{ "requestId": "ocx-f588fccd24993ea85e2df4cf99722700", "timestamp": 1790840465696, "model": "Qwen3.8-Flash-Next", "provider": "AMD", "surface": "claude", "admissionKind": "loopback", "inboundProtocol": "messages", "accountLogLabel": "kb697860e8b4c1dd0df9938d2160aeca1", "conversationId": "11d36ad5d7886b58c5856814aa80018b", "requestedModel": "AMD/Qwen3.8-Flash-Next", "requestedAlias": "AMD/Qwen3.8-Flash-Next", "requestedEffort": "xhigh", "effectiveEffort": "xhigh", "reasoningWireField": "reasoning_effort", "reasoningWireValue": "xhigh", "configuredServiceTier": "fast", "configuredSpeedLabel": "fast", "modelSupportsServiceTier": false, "tierOutcome": { "wireKind": null, "wireValue": null, "fastOutcome": "unknown", "confirmation": "unknown" }, "resolvedModel": "Qwen3.8-Flash-Next", "servedModel": "Qwen3.8-Flash-Next", "status": 502, "durationMs": 1917, "errorCode": "upstream_server_error", "terminalStatus": "failed", "closeReason": "terminal", "upstreamError": "System message must be at the beginning.", "usageStatus": "unreported", "attempts": [ { "ordinal": 1, "provider": "AMD", "model": "Qwen3.8-Flash-Next", "adapter": "openai-chat", "status": 502, "durationMs": 1904, "sendCount": 1, "recoveryKinds": [], "usageStatus": "unreported", "requestedEffort": "xhigh", "effectiveEffort": "xhigh", "reasoningWireField": "reasoning_effort", "reasoningWireValue": "xhigh", "tierOutcome": { "wireKind": null, "wireValue": null, "fastOutcome": "unknown", "confirmation": "unknown" }, "accountLogLabel": "kb697860e8b4c1dd0df9938d2160aeca1", "deliverySummary": { "adapterEvents": 1, "relayedEvents": 3, "semanticBytes": 0, "sideEffectEvents": 0, "terminalEvents": 1 }, "errorCode": "upstream_server_error", "failureStage": "protocol-prelude", "failureCause": "upstream-fault", "resendPermission": "refused-ambiguous", "displayMetrics": { "tokPerSecond": { "kind": "unavailable", "reason": "usage_missing" }, "decodeTokPerSecond": { "kind": "unavailable", "reason": "usage_missing" }, "cost": { "kind": "unavailable", "reason": "usage_missing" } } } ], "spend": { "sends": 1, "settled": 1, "unresolved": 0 }, "transportPhase": "terminal_sse", "terminalSource": "upstream", "routeDecision": { "version": 1, "decisionId": "552a2bae5447", "createdAt": 1790840465708, "requestedModel": "AMD/Qwen3.8-Flash-Next", "routeKind": "explicit-provider", "requirements": [], "candidates": [ { "provider": "AMD", "model": "Qwen3.8-Flash-Next", "eligible": true, "exclusions": [] } ], "selected": { "candidateIndex": 0, "provider": "AMD", "model": "Qwen3.8-Flash-Next", "reason": "explicit-provider-namespace" } }, "protocolTrace": { "v": 1, "inbound": "messages", "mode": "legacy-bridge", "upstream": "chat", "requestPath": [ "messages", "responses-internal", "ir", "chat" ], "responsePath": [ "chat", "ir", "responses-internal", "messages" ], "reasonCodes": [ "not-migrated" ], "featureEffects": [ { "feature": "request.tools", "disposition": "translated" }, { "feature": "request.reasoning", "disposition": "degraded" }, { "feature": "request.cache_control", "disposition": "degraded" } ], "attempts": [ { "ordinal": 1, "upstream": "chat", "mode": "legacy-bridge", "requestPath": [ "messages", "responses-internal", "ir", "chat" ] } ], "contractVersion": "2026-09-25.2" }, "failureStage": "protocol-prelude", "failureCause": "upstream-fault", "resendPermission": "refused-ambiguous", "displayMetrics": { "tokPerSecond": { "kind": "unavailable", "reason": "usage_missing" }, "decodeTokPerSecond": { "kind": "unavailable", "reason": "usage_missing" }, "cost": { "kind": "unavailable", "reason": "combo_attempt_unavailable" } } }github-actions commentedon Oct 1, 2026
Automated translation bookkeeping — detected language: English.
fix(openai-chat): handle Flash-Next leading-system templates (#6966)