"Provider returned error" with openrouter/openai/gpt-5.1-codex

Description

Here's the error from above with --print-logs:

https://gist.github.com/jdanbrown/bf967e266b071a7651e4aead06ee2990

Here's a formatted version of the big error={...} json blob from --print-logs:

https://gist.github.com/jdanbrown/e2113851b6c65834511b31fd53f88a7a

Here's my config:

$ opencode debug config
{
  "$schema": "https://opencode.ai/config.json",
  "theme": "jdanbrown",
  "keybinds": {
    "leader": "ctrl+o",
    "app_exit": "ctrl+d,<leader>q",
    "editor_open": "<leader>e",
    "theme_list": "<leader>t",
    "sidebar_toggle": "<leader>b",
    "status_view": "<leader>s",
    "session_export": "<leader>x",
    "session_new": "<leader>n",
    "session_list": "<leader>l",
    "session_timeline": "<leader>g",
    "session_share": "none",
    "session_unshare": "none",
    "session_interrupt": "ctrl+c",
    "session_compact": "<leader>c",
    "messages_page_up": "pageup",
    "messages_page_down": "pagedown",
    "messages_half_page_up": "ctrl+alt+u",
    "messages_half_page_down": "ctrl+alt+d",
    "messages_first": "ctrl+alt+a",
    "messages_last": "ctrl+alt+e",
    "messages_copy": "<leader>y",
    "messages_undo": "<leader>u",
    "messages_redo": "<leader>r",
    "messages_toggle_conceal": "<leader>h",
    "model_list": "<leader>m,ctrl+j",
    "model_cycle_recent": "f2",
    "model_cycle_recent_reverse": "shift+f2",
    "command_list": "ctrl+_",
    "agent_list": "<leader>a",
    "agent_cycle": "tab",
    "agent_cycle_reverse": "shift+tab",
    "input_clear": "ctrl+c",
    "input_forward_delete": "ctrl+d",
    "input_paste": "ctrl+v",
    "input_submit": "return",
    "input_newline": "shift+enter,alt+enter",
    "history_previous": "up,ctrl+p",
    "history_next": "down,ctrl+n",
    "session_child_cycle": "<leader>right",
    "session_child_cycle_reverse": "<leader>left"
  },
  "tui": {
    "scroll_speed": 2
  },
  "share": "manual",
  "autoupdate": false,
  "model": "openrouter/anthropic/claude-haiku-4.5",
  "small_model": "openrouter/google/gemini-2.5-flash",
  "provider": {
    "openrouter": {
      "models": {
        "anthropic/claude-sonnet-4.5@thinking-32k": {
          "id": "anthropic/claude-sonnet-4.5",
          "options": {
            "reasoning": {
              "effort": "high"
            }
          }
        }
      }
    }
  },
  "mcp": {
    "context7": {
      "type": "remote",
      "url": "https://mcp.context7.com/mcp",
      "headers": {
        "CONTEXT7_API_KEY": ...
      }
    }
  },
  "formatter": {
    "swiftformat": {
      "command": [
        "swiftformat",
        "$FILE"
      ],
      "extensions": [
        ".swift"
      ]
    }
  },
  "instructions": [],
  "permission": {
    "bash": {
      "*": "ask",
      "git status": "allow",
      "git diff": "allow",
      "git log": "allow"
    }
  },
  "agent": {},
  "mode": {},
  "plugin": [],
  "command": {},
  "username": "danb"
}

OpenCode version

1.0.85

Steps to reproduce

(above)

Screenshot and/or share link

No response

Operating System

macOS 15.5

Terminal

VS Code terminal

Same error if I comment out the "mcp": { ... } block in my config.

Same error with empty config, too.

Same error with openrouter/openai/gpt-5.1-codex-mini.

Does not error with openai/gpt-5.1-codex (i.e. without openrouter).

Hmmm

I have a fix for this, that's really weird tho I think the openrouter ai sdk has a bug maybe

Maybe related to the gemini-3 openrouter issue?

"Provider returned error" with openrouter/openai/gpt-5.1-codex #4608 — this issue
Incomplete response with openrouter/google/gemini-3-pro-preview #4615 — gemini-3 issue

yeah it's related to their ai sdk doing something

or we are doing something wrong w it

Talked w/ openrouter they are fixing it

I’ve been running into the same issue when using OpenCode with OpenRouter + OpenAI models (e.g. openrouter/openai/gpt-5.1-codex).

After digging into the logs and request payload, I confirmed that the root cause is:

OpenCode re-sends prior "assistant" messages including the reasoning / reasoning_details metadata
When OpenRouter forwards that to OpenAI, the encrypted reasoning blocks (providerMetadata with reasoning.encrypted items like rs_...) no longer match the OpenAI project/key OpenRouter is using
OpenAI then rejects the request with:

The encrypted content for item rs_0aa9cdb85cd23c86016920cad04e4c8196bf0bdb7dbce6a4df could not be verified.
type: invalid_request_error

To work around this, I patched my local OpenCode clone to strip reasoning provider metadata only when using the OpenRouter provider.

Concretely, I added an option to MessageV2.toModelMessage(...):

export function toModelMessage(
  input: { info: Info; parts: Part[] }[],
  options?: {
    // When true, strips reasoning providerMetadata to avoid OpenAI encrypted content verification errors via OpenRouter
    stripReasoningMetadata?: boolean
  },
): ModelMessage[] { ... }

and when model.npm === "@openrouter/ai-sdk-provider", I call it with stripReasoningMetadata: true.

Inside toModelMessage, for assistant messages I now do (simplified):

if (part.type === "text")
  assistantMessage.parts.push({
    type: "text",
    text: part.text,
    providerMetadata: options?.stripReasoningMetadata ? undefined : part.metadata,
  })

if (part.type === "tool") {
  ...
  callProviderMetadata: options?.stripReasoningMetadata ? undefined : part.metadata,
}

if (part.type === "reasoning") {
  assistantMessage.parts.push({
    type: "reasoning",
    text: part.text,
    // Skip providerMetadata when stripReasoningMetadata is true to avoid
    // OpenAI encrypted content verification errors when using OpenRouter
    providerMetadata: options?.stripReasoningMetadata ? undefined : part.metadata,
  })
}

So the encrypted reasoning blocks are no longer forwarded when going through OpenRouter, but everything else (including the reasoning text itself) is preserved. Other providers (direct OpenAI, Anthropic, etc.) keep the original behavior.

With this patch in place, I can now use openrouter/openai/gpt-5.1-codex in OpenCode without hitting the “Provider returned error / encrypted content could not be verified” issue.

If it’s helpful, I’m happy to turn this into a PR — or feel free to adapt this approach in whatever way best fits the existing provider abstractions.

if you update to latest this should be fixed, may need to clear your cache too, try: opencode models --refresh

👍 Thanks! I got through one non-trivial prompt without issues with openrouter/openai/gpt-5.1-codex. I'll bang on it more tomorrow.

for people in sanctioned regions (Russia, Iran, etc. i don't know them all) try running it through proxy. i thought issue persisted but it was in fact a region restriction

I appear to be having the same from my vultr vps but non gpt models from the same openrouter seem to be working?

This should be fixed in the next release (1.1.24) by commit: 40836e9

Yep, appears to now be fixed in 1.1.25:

Did you also fix this other issue in the process? I just confirmed that my repro from the top of it is now working for me in 1.1.25:

"Provider returned error" with openrouter/openai/gpt-* #7542

Navigation Menu

Repository navigation

"Provider returned error" with openrouter/openai/gpt-5.1-codex #4608

Description

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

"Provider returned error" with openrouter/openai/gpt-5.1-codex #4608

Description

Description

OpenCode version

Steps to reproduce

Screenshot and/or share link

Operating System

Terminal

Activity

jdanbrown commented on Nov 21, 2025

jdanbrown commented on Nov 21, 2025

jdanbrown commented on Nov 21, 2025

jdanbrown commented on Nov 21, 2025

rekram1-node commented on Nov 21, 2025

rekram1-node commented on Nov 21, 2025

jdanbrown commented on Nov 21, 2025

rekram1-node commented on Nov 21, 2025

rekram1-node commented on Nov 21, 2025

rekram1-node commented on Nov 21, 2025

jhkim2goqual commented on Nov 23, 2025

rekram1-node commented on Nov 24, 2025

jdanbrown commented on Nov 25, 2025

JohnBakhmat commented on Dec 17, 2025

NathanRHD commented on Jan 10, 2026

rekram1-node commented on Jan 16, 2026

jdanbrown commented on Jan 16, 2026

jdanbrown commented on Jan 16, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions