freellmpool › providers › Cerebras

Cerebras API: fast inference with finite trial credit

Cerebras runs inference on its wafer-scale engine, giving some of the highest tokens-per-second available. Its current pricing offers a finite $5 trial rather than recurring free daily capacity, so freellmpool keeps the documented Cerebras routes out of automatic routing. Get a key at cloud.cerebras.ai and call the OpenAI-compatible endpoint at https://api.cerebras.ai/v1. Use freellmpool to explicitly pin Cerebras and fall back to recurring tiers for normal traffic.

What the Cerebras trial is good for

Pick Cerebras when you want to evaluate its throughput within a bounded trial budget. Its documented shared catalog is deliberately small. It is a useful explicit experiment, not a source of automatic or recurring-free pool capacity.

Documented trial models

Cerebras' current shared model catalog lists:

• gpt-oss-120b — strong open reasoning model, served fast.
• gemma-4-31b — Google's current Gemma route.

zai-glm-4.7 is no longer on Cerebras' official shared model list and is disabled in freellmpool.

Get a key and call it

curl https://api.cerebras.ai/v1/chat/completions \
  -H "Authorization: Bearer $CEREBRAS_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"gpt-oss-120b","messages":[{"role":"user","content":"Hi"}]}'

Limits and gotchas

• The $5 trial is finite. Check current pricing and do not treat an old requests-per-day figure as a recurring allowance.
• The model list is intentionally small; don't expect dozens of variants.
• gpt-oss is a reasoning model — give it enough max_tokens so hidden reasoning doesn't eat the whole budget and return empty content.

Pool Cerebras with other free tiers

Keep Cerebras outside automatic routing so the finite trial is intentional. freellmpool can explicitly pin a Cerebras model while recurring providers handle ordinary traffic:

pip install freellmpool
export CEREBRAS_API_KEY=...              # plus other free keys
freellmpool ask -m cerebras/gpt-oss-120b "..."  # explicitly spend trial credit
freellmpool benchmark                    # time each provider you've configured

See also Groq (a recurring fast tier), best free LLM API gateway, and using multiple free LLM APIs together.

FAQ

Is the Cerebras API free?

Cerebras currently offers a finite $5 trial rather than recurring free capacity. Get a key at cloud.cerebras.ai, review current pricing, and explicitly pin a documented model.

Is Cerebras faster than Groq?

Both are very fast and often trade the top spot depending on model and load. The practical answer is to run freellmpool benchmark and let routing prefer whichever is fastest for you right now.

Why is Cerebras excluded from automatic routing?

Automatic fan-out could consume finite trial credit unexpectedly. Explicit pins keep that spend intentional while still making the fast endpoint available.

Part of freellmpool (MIT, open source). Limits change — check Cerebras' docs. Updated 2026-08-29.