Skip to content

Big Pickle (big-pickle) model returns AI_APICallError - was working before #28141

New issue
New issue
Closed as not planned
Closed as not planned
Big Pickle (big-pickle) model returns AI_APICallError - was working before#28141
Assignees
MrMushrooooom

Description

@mohammeditegypt-dot
mohammeditegypt-dot
opened on May 18, 2026

Description

The Big Pickle (big-pickle) model from OpenCode Zen has stopped responding.
I was able to use it successfully before (completed a full application), but
as of today (May 18, 2026), it no longer works. Other free models like
DeepSeek V4 Flash Free work perfectly.

Plugins

No response

OpenCode version

1.15.4

Steps to reproduce

  1. Open OpenCode Desktop (latest version, Windows 11)
  2. Select model: Big Pickle (big-pickle)
  3. Send any message
  4. The model does not respond / returns AI_APICallError

Screenshot and/or share link

ERROR service=llm providerID=opencode modelID=big-pickle error={"error":{"name":"AI_APICallError", "url":"https://opencode.ai/zen/v1/messages", "requestBodyValues":{"model":"big-pickle"}}}

  • I cleared the cache at %USERPROFILE%\.cache\opencode and restarted — no change
  • The config file at %USERPROFILE%\.config\opencode\opencode.jsonc has no custom provider settings (only $schema)
  • Big Pickle appears in the Zen free models list and is selectable, but API calls fail
  • Other free models (DeepSeek V4 Flash Free, Qwen3.6 Plus Free) work without issues on the same setup

Operating System

windows 10

Terminal

No response

Activity

  1. github-actions
    assigned
    MrMushrooooom
    on May 18, 2026

  2. github-actions commented on May 18, 2026

    @github-actions
    github-actionsBot
    on May 18, 2026 – with GitHub Actions
    Contributor

    This issue might be a duplicate of an existing issue. Please check:

    • Big Pickle not working. #28138: Same model (big-pickle) failing on the same version (1.15.4) with similar symptoms after upgrading — may be the same underlying regression.

  3. aentrepreneur commented on Jul 9, 2026

    @aentrepreneur
    aentrepreneur
    on Jul 9, 2026

    Diagnostic from opencode/big-pickle logs (v1.17.16, Linux Ubuntu 24.04)

    Config

    { "model": "opencode/big-pickle" }

    Error timeline for big-pickle (Jun 11 – Jul 9)

    Error Frequency Period
    Rate limit exceeded 60+ occurrences Jun 11 – Jul 9
    Socket connection closed 20+ Jun 21
    Prima Labs: Invalid model deepseek-ai/DeepSeek-V4-Flash 8 Jul 9 (today)
    Internal Server Error 4 Jun 24
    Service Unavailable 2 Jun 23, Jul 4
    Cannot connect to API 10+ Jun 13, Jun 23
    Insufficient Balance 2 Jul 3

    🔍 New diagnostic: exact root cause from Jul 9 logs

    The Prima Labs error started today and provides the exact model routing:

    AI_APICallError: Error from provider (Prima Labs): /chat/completions:
Invalid model name passed in model=deepseek-ai/DeepSeek-V4-Flash.
Call `/v1/models` to view available models for your key.

    The OpenCode routing sends opencode/big-pickle → Prima Labs provider with model name deepseek-ai/DeepSeek-V4-Flash, which does not exist on Prima Labs' API.

    Key differences from #28138

    #28138 (May 18) Our logs (Jul 9)
    Error "Model big-pickle not supported for format anthropic" "deepseek-ai/DeepSeek-V4-Flash not a valid model on Prima Labs"
    Nature Format mapping bug Invalid model name on upstream provider

    This could be:

    1. A new routing configuration pushing big-pickle through Prima Labs with the wrong model name
    2. Prima Labs dropping/deprecating that model
    3. A regression in the routing layer post internal refactor

    Suggest checking whether deepseek-ai/DeepSeek-V4-Flash is still a valid model on Prima Labs (call /v1/models) and whether the routing should be updated to a valid model or a different provider.

  4. aentrepreneur commented on Jul 27, 2026

    @aentrepreneur
    aentrepreneur
    on Jul 27, 2026

    Update (Jul 27, 2026) — v1.18.4 → v1.18.7: Bug persists

    Environment

    • OS: Ubuntu 24.04 (WSL2 on Windows 11)
    • OpenCode: v1.18.4 (Jul 22-24) → v1.18.7 (Jul 27)
    • Model: opencode/big-pickle
    • Config: Minimal — only "model": "opencode/big-pickle" + MCP servers. No custom provider settings.

    Live endpoint test (just performed)

    Endpoint Status Latency Notes
    GET /v1/models 200 OK 0.39s big-pickle appears in list
    POST /chat/completions (big-pickle) 200 OK 4.3s content: null, only reasoning tokens, finish_reason: length

    The /v1/models endpoint works fine — no general outage. The /chat/completions endpoint occasionally succeeds but returns empty content.

    Error timeline (last 5 days from logs)

    Date OpenCode version Internal 500 errors Other errors
    Jul 22 1.18.4 34 —
    Jul 23 1.18.4 64 3 rate limit
    Jul 24 1.18.4 172 —
    Jul 25 — 0 (no session)
    Jul 26 — 0 (no session)
    Jul 27 1.18.7 39 1 rate limit, 1 socket closed

    Total across all time (Jun 10 — Jul 27):

    • Internal server error: 315
    • Rate limit exceeded: 96
    • Socket connection closed: 46
    • Prima Labs invalid model name: 18
    • Service Unavailable: 10
    • Unauthorized: 25

    Burst pattern (Jul 24, peak day)

    Errors fire in rapid bursts of 2-4 seconds between retries, indicating the backend is consistently rejecting requests:

    13:11:23 → 13:11:26 → 13:12:01 → 13:12:03 → 13:12:08 → 13:12:23 → 13:12:26
13:18:40 → 13:19:34 → 13:19:39 → 13:19:44 → 13:19:53
14:11:52 → 14:11:54 → 14:12:35 → 14:12:37 → 14:12:42

    Peak minute: 7 errors in a single minute (15:37).

    Agents affected

    Agent Errors (last 5 days)
    build 187
    plan 89
    compaction 16
    general 4

    Key observations

    1. Bug persists across versions: Same Internal server error on v1.18.4 (Jul 22-24) and v1.18.7 (Jul 27). Not a client-side regression.
    2. Sporadic success: The endpoint occasionally returns 200 but with content: null — the model is listed in /v1/models but /chat/completions is unstable.
    3. Only big-pickle affected: Other models (deepseek, qwen, kimi-k2.6) on the same setup barely fail.
    4. Backend issue: The error is server-side (500 from OpenCode Zen), not a client misconfiguration.

    Request

    Is there a plan to stabilize the big-pickle backend routing? The model has been unreliable since at least May 18 (when this issue was opened) and the 500 errors continue through v1.18.7.

    Related issues

  5. aentrepreneur commented on Aug 4, 2026

    @aentrepreneur
    aentrepreneur
    on Aug 4, 2026

    Additional Evidence: Comprehensive Log Analysis (v1.18.11 - Linux)

    I can confirm this issue persists on OpenCode v1.18.11 running on Ubuntu 24.04 (WSL2). Below is a comprehensive analysis of global OpenCode logs spanning June-August 2026.

    Error Summary

    Error Type Count Description
    AI_APICallError: Rate limit exceeded ~50+ Exceeded requests per minute/hour
    AI_APICallError: Internal Server Error ~28 Server-side errors from OpenCode backend
    AI_APICallError: Cannot connect to API ~96 Intermittent connectivity issues
    AI_APICallError: Monthly usage limit reached 2+ Monthly usage cap hit
    Failed to fetch models.dev 4 Model catalog fetch failure

    Model Error Distribution

    Model Stream Errors % of Total
    opencode/big-pickle 627 95.2%
    openai/gpt-5.4-fast 32 4.9%
    All other models 25+ <1%

    Key Finding: big-pickle accounts for 95%+ of all stream errors across the entire log history.

    Agent/Mode Impact

    Agent Errors Mode
    build 387 primary
    plan 263 primary
    compaction 29 primary
    title 13 primary

    Worst Affected Sessions

    Session ID Total Errors Active Period
    ses_0b693d445ffemR5Z5MW4bgAOfk 304 Jul 10 - Aug 4
    ses_079db9c5cffeyMA5D4s5pkXUCP 40 -
    ses_05bf1a8ceffei45iKFaJHH8xfP 35 -
    ses_14183c4e6ffesPpG5kDroz4ERK 33 Jun 13

    Error Timeline (Daily Peaks)

    Date Errors
    2026-07-24 182
    2026-07-27 135
    2026-07-23 77
    2026-07-22 58
    2026-06-13 52
    2026-08-04 29

    Specific Error Patterns Found

    1. Rate Limit Errors (Jun 11, 2026)

    2026-06-11T18:00:56 → "Rate limit exceeded. Please try again later."
2026-06-11T18:04:22 → "Failed after 3 attempts. Last error: Rate limit exceeded."

    2. Monthly Usage Limit (Jun 11, 2026)

    2026-06-11T19:50:56 → "Monthly usage limit reached. Resets in 14 days."
                        "To continue using this model now, enable usage from your available balance:"
                        "https://opencode.ai/workspace/wrk_01KMNPSZN5HAM8K0HAECRK5W8G/go"

    3. Internal Server Error Cluster (Aug 4, 2026)

    2026-08-04T03:05:06 → AI_APICallError: Internal Server Error (plan mode)
2026-08-04T03:05:10 → AI_APICallError: Internal Server Error (plan mode)
2026-08-04T03:05:15 → AI_APICallError: Internal Server Error (plan mode)
... (28 total in this cluster, spanning 03:05-03:41 UTC)

    4. Connectivity Failures (Jun 13, 2026)

    2026-06-13T01:17:44 → "Cannot connect to API: Unable to connect. Is the computer able to access the url?"
2026-06-13T01:17:46 → "Cannot connect to API: Was there a typo in the url or port?"

    These errors also affected gpt-5.4-fast on the same day, suggesting a broader backend issue.

    Configuration

    • Primary Model: opencode/big-pickle (configured as default)
    • OpenCode Version: 1.18.11
    • Platform: Ubuntu 24.04 (WSL2)
    • Active Accounts: OpenAI, Groq, OpenRouter, opencode-go, HuggingFace

    Comparison with Other Free Models

    Other free models (DeepSeek V4 Flash, Qwen3.6 Plus) work without issues on the same setup, as reported in the original issue. This confirms the problem is specific to big-pickle.

    Conclusion

    The issue is not connectivity-related (not DNS, firewall, or local config). The root cause appears to be:

    1. Rate limiting on the OpenCode Zen backend for big-pickle
    2. Monthly usage caps that are not clearly communicated
    3. Intermittent Internal Server Errors from the backend

    The high error count (627 stream errors for big-pickle alone) suggests either:

    • The free tier for this model is severely degraded
    • There is a backend stability issue specific to this model
    • Rate limit thresholds are set too low for typical usage patterns

    Environment: Ubuntu 24.04 | OpenCode v1.18.11 | WSL2
    Reporter: Angelo Esquivel (CyberSecurity)

  6. Aymen104 commented on Aug 4, 2026

    @Aymen104
    Aymen104
    on Aug 4, 2026

    Big Pickle intermittent failures — hang then cancel (v1.18.12)

    Big Pickle (opencode/big-pickle) intermittently fails with hang-then-cancel behavior. Message sends, model appears to process, then request aborts or returns [400] Provider returned error.

    Environment:

    • OpenCode: v1.18.12
    • OS: Windows 10 x64
    • Model: opencode/big-pickle

    Error logs:

    [15:08:16] SESSION ERROR ... Aborted
[15:26:23] SESSION ERROR ... Error from provider (Console): Upstream request failed: [400] Provider returned error
[15:34:43] SESSION ERROR ... Aborted
[15:35:20] SESSION ERROR ... Error from provider (Console): Upstream request failed: [400] Provider returned error

    Additional info:

    • Started today (Aug 4, 2026), was working fine before
    • Other free models (DeepSeek V4 Flash Free, MiMo V2.5 Free) work without issues
    • Direct API test to https://opencode.ai/zen/v1/chat/completions with model big-pickle returns successful response with reasoning + content
    • Model appears to be MiMo-v2.5 underneath (response content says "I'm MiMo-v2.5")
    • Added explicit provider config with "npm": "@ai-sdk/openai-compatible" — did not resolve the issue
    • Related to Big Pickle (big-pickle) model returns AI_APICallError - was working before #28141 and Big Pickle not working. #28138

    Steps to reproduce:

    1. Set model to opencode/big-pickle
    2. Send messages
    3. Sometimes works, sometimes hangs then cancels

    Note: This is NOT a config issue. The API works when called directly. The problem is in OpenCode's internal routing/provider handling for Big Pickle.

  7. github-actions commented on Oct 5, 2026

    @github-actions
    github-actionsBot
    on Oct 5, 2026 – with GitHub Actions
    Contributor

    To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.

  8. github-actions
    closed this as not plannedon Oct 5, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions