You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Change default CEREBRAS_MODEL — gpt-oss-120b is gated on most tiers #3
gpt-oss-120b is gated behind specific Cerebras tiers. Users on the free tier or lower paid tiers get 404s when the MCP tries to call it, which manifests as the plugin failing to generate anything despite being connected.
Evidence
Tested against Cerebras /models and /chat/completions endpoints on the free tier:
Model
/models
/chat/completions
qwen-3-235b-a22b-instruct-2507
✅
✅ works
llama3.1-8b
✅
✅ works
gpt-oss-120b (current default)
✅
❌ 404 gated
zai-glm-4.7
✅
❌ 404 gated
The /models endpoint lists all four but only two are actually callable on this tier.
Proposed fix
Change the default in cerebras-client.ts to qwen-3-235b-a22b-instruct-2507 — it's the strongest of the two reliably available models for codegen.
Alternative: make the client probe /models on first call and pick the best available from a preference list. More robust but more code.
Workaround until this is fixed
Set CEREBRAS_MODEL=qwen-3-235b-a22b-instruct-2507 in ~/.claude/settings.json env block.
Note
Documentation in CLAUDE.md should also mention that the default may not work on all tiers and recommend setting CEREBRAS_MODEL explicitly.
Activity
jiffy-2389 commentedon Oct 27, 2026
❌ Jiffy stopped
Comment
/jiffy retryto run it again.Recent stages
harperreed commentedon Oct 28, 2026
/jiffy retry