Description

Multiple free models in OpenCode v1.14.40 intermittently fail during generation, returning a generic error:

“Provider did not respond”

This behavior appears under specific conditions including long generations, repeated usage, and modified reasoning parameters. The issue is reproducible across multiple free-tier models and impacts stability of responses.

Affected Models

• Nemotron 3 Super Free (primary and most consistent failure case)
• GPT-5 Nano (observed after extended usage sessions)
• Other free-tier models (inconsistent but recurring reports)

Environment

• OpenCode version: v1.14.40
• Operating System: Windows 11 Pro (Version 10.0.26200, Build 26200)
• Terminal: Windows Terminal (Desktop)
• Plugins: None

Steps to Reproduce

Case 1: Nemotron 3 Super Free (mid-generation failure)

1. Open OpenCode v1.14.40 on Windows 11

2. Select Nemotron 3 Super Free

3. Send a moderately long prompt

4. Allow full generation to begin

5. Observe failure mid-response with:

   Provider did not respond

Case 2: Reasoning parameter instability

1. Select Nemotron 3 Super Free or similar free model

2. Modify reasoning parameters from default settings (any non-default configuration)

3. Submit prompt

4. Observe either:

   • No response
   • Immediate “Provider did not respond” error

Case 3: GPT-5 Nano session degradation

1. Use GPT-5 Nano continuously for multiple prompts (extended session usage)

2. After several successful responses (5–15+ interactions)

3. Submit another prompt

4. Observe failure:

   Provider did not respond

Expected Behavior

• Model should consistently return complete responses
• Reasoning parameter changes should not affect availability or connectivity
• Long sessions should not degrade into provider failure states

Actual Behavior

• Intermittent mid-generation termination
• Immediate request failure when reasoning parameters are modified
• Session-based degradation leading to temporary loss of model availability

Notes / Observations

• Error is not tied to a single prompt type; both short and long prompts can fail, but longer outputs increase likelihood.
• Retry sometimes succeeds immediately, suggesting transient provider instability rather than deterministic crash.
• Issue appears more frequent on free-tier models compared to paid or stable endpoints.
• No plugin involvement.

Impact

• Disrupts reliability of free-tier model usage
• Breaks long-form generation workflows
• Reduces usability when experimenting with reasoning settings
• Causes inconsistent developer experience in OpenCode environment