OpenCode Go API request failed (400)

[itsmorty]

Type: Bug

Request Logs:

[2026-09-22T08:13:09.128Z] Request: initiator=github.copilot-chat model=opencodego:deepseek-v4.1-flash::session-2026-05-21-b rawModel=deepseek-v4.1-flash endpoint=chat-completions metadataSource=models.dev messages=168 promptEstimate=184701 maxOutputTokens=384000 session=vscode-0f4a3f3c request=req-33226406 modelConfiguration={"reasoningEffort":"low"} thinkingSource=modelConfiguration thinking={"deepseek":"low","glm":"off","kimi":"off","minimax":"off","openai":"off","qwen":"off","qwenBudget":"auto","mimo":"off","muse":"off"} thinkingPayload={"reasoning_effort":"low"}
[go-gw] model=deepseek-v4.1-flash hasReasoningEffort=true treatReasoningAsContent=false
[request] url=https://opencode.ai/zen/go/v1/chat/completions payloadBytes=694636 requestTimeoutMs=600000 streamIdleTimeoutMs=120000
[http-error-body] {"error":{"param":null,"type":"invalid_request_error","code":"invalid_request_error","message":"Upstream request failed: [invalid_request_error] The `reasoning_content` in the thinking mode must be passed back to the API."}}
[http] 400 Bad Request content-type=application/json

How to see this error:

I used the "deepseek-v4.1-flash" model with the reasoning effort "low". If the error happens, it retries 3 times and fails. You can hit the "try again" button, but it keeps giving the same error. If you send a prompt, it continues without an error.

Extension version: 0.7.5
VS Code version: Code 1.137.0 (645f29cc3176500b4b5762ba887cf2a7f0ffdf2c, 2026-09-08T13:43:59-07:00)
OS version: Linux x64 7.2.6-arch2-1
Modes:

System Info

Item	Value
CPUs	AMD Ryzen 7 7435HS (16 x 2528)
GPU Status	2d_canvas: enabled GPU0: VENDOR= 0x10de [Google Inc. (NVIDIA Corporation)], DEVICE=0x28e1 [ANGLE (NVIDIA Corporation, NVIDIA GeForce RTX 4050 Laptop GPU/PCIe/SSE2, OpenGL ES 3.2 NVIDIA 615.71.09)], DRIVER_VENDOR=NVIDIA, DRIVER_VERSION=615.71.09 ACTIVE Machine model name: Machine model version: direct_rendering_display_compositor: disabled_off_ok gpu_compositing: enabled multiple_raster_threads: enabled_on opengl: enabled_on rasterization: enabled raw_draw: disabled_off_ok skia_graphite: disabled_off trees_in_viz: disabled_off video_decode: enabled video_encode: disabled_software vulkan: enabled_on webgl: enabled webgpu: enabled webgpu_on_vk_via_gl_interop: enabled webnn: disabled_off
Load (avg)	2, 3, 2
Memory (System)	31.15GB (14.41GB free)
Process Argv	--ozone-platform=wayland --password-store=gnome-libsecret --crash-reporter-id 10b49d8a-5d3e-4007-9569-c651e5777184
Screen Reader	no
VM	0%
DESKTOP_SESSION	plasma
XDG_CURRENT_DESKTOP	KDE
XDG_SESSION_DESKTOP	KDE
XDG_SESSION_TYPE	wayland

A/B Experiments

vsliv368cf:30146710
binariesv615:30325510
nativeloc1:31344060
dwcopilot:31170013
dwoutputs:31242946
copilot_t_ci:31333650
e5gg6876:31282496
pythonrdcb7:31342333
6518g693:31463988
aj953862:31281341
envsactivate1:31551504
editstats-enabled:31346257
478i5457:31544283
cloudbuttont:31379625
3efgi100_wstrepl:31403338
ec5jj548:31422691
cp_cls_t_966_ss:31526232
4je02754:31466945
8hhj4413:31478653
cp_cls_c_1081:31454833
conptydll_true:31498968
e9c30283:31461165
46204921:31447328
ei9d7968:31496641
chat:31457767
8hig5102:31480529
89g7j272:31518289
i2gc6536:31499202
52612955:31516516
githubstableflight:31469463
ddid_c:31478207
hmra_i5g22:31518061
7df3h592:31512476
cp_cls_t_1082:31535311
logging_enabled_new:31498466
j0d79568:31499440
jb_cp_cls_t_632:31543129
56dj4588:31512888
32d76977:31512328
ha629193:31508444
jh5f2457_c:31540921
jbcp_cls_pctr_t:31531130
cp_intellij_t_nes:31548657
ahp-both-windows:31556933
ihg5j128:31534457
9gg7f176:31542112
sandbox-ui-on-2:31543161
nes-ex-cf-euporie-03:31542868
enable_editor_pane_layout:31569726
allow-none:31555437
ac562147:31564510
c7c27ce7:31554789
1h923230:31564177
treatment-23-1:31555779
unuse_dynamic_mcp:31555281
0d8dfbc3:31570291
0c1h4866:31566224
session-mark-done:31558131
5b8j3302:31564601
autotiers:31584978
mangle-name-control:31572633
allow-signed-out:31584137
intellij_nes_preview_t:31580086
39a5d156:31574986
0ec91c74:31585910
unread:31581112
links-treatment:31587702
aw-autom-on:31583758
revealsession_control_7ed7e840:31582193
usage_tips_endabled_0911:31587631
7g328426_fixed:31584544
sessioninlinediffenabled_disabled:31587634
permission_prompt_treatment:1332566
ccr_pr_nudge_control:1319470
vsc_wsm_c:1340633
g03gg798:1417072
cfee5631:1379228
ihko4107:1412285
3aced641:1389836
jc553158:1416789

[ltmoerdani]

@itsmorty Thanks for the detailed report, the request logs made this one trackable.

Short version: the reasoning echo itself was working, but when Copilot Chat summarizes or compacts a long conversation (or history trimming cuts the wrong turn), the reasoning_content from earlier turns disappears from what we replay. DeepSeek's thinking mode then rejects every follow-up request until a brand new conversation, and retrying can't help because the history shape never changes.

Fixed in #242: when the gateway returns this exact error, we now retry once with the stale reasoning echo stripped and reasoning_effort turned off, so the request is self-contained again and the error stops recurring for the rest of the conversation. Covered by a mock-server e2e that reproduces the validator behavior, plus regression tests.

Will ship in the next release. Thanks again for the logs, they pinpointed it.