Cobble hosts a curated catalog of open-source models, quantized to FP8, served on reclaimed hardware. The catalog is the single source of truth — what /v1/models returns is exactly what the API accepts.
Discover models programmatically
bash
curl https://api.cobble.network/v1/models \
-H "Authorization: Bearer $COBBLE_API_KEY"Every agent and SDK that supports OpenAI-compatible providers can read this endpoint. Model IDs are stable — we never rename a published ID.
Recommended defaults
Don't want to evaluate the whole catalog? Start here:
| You're doing | Use | Why |
|---|---|---|
| Agentic coding | deepseek/deepseek-v4-flash | Strongest tool-calling and repo-scale reasoning in the catalog |
| Coding on a budget window | qwen/qwen3.6-35b-a3b | MoE efficiency — near-flagship coding at a fraction of the compute |
| General agents / tool use | qwen/qwen3.5-122b-a10b | Flagship-class reasoning, 128K context |
| Writing, chat, roleplay | gemma4/gemma4-26b-a4b | Best prose quality per request on the Creative plan |
| Fast and cheap everything | mistral/mistral-nemo-12b | Lowest latency in the catalog |
| Embeddings | nomic/nomic-embed-1.5 | Billed from wallet at $0.01 / 1M tokens |
| Document OCR | chandra/chandra-ocr | Billed from wallet, per 1K pages |
Model IDs above are drafts — sync with
/v1/modelsbefore publishing.
Which models come with which plan
Each plan includes everything in the plans below it.
- Creative — small/fast tier: Mistral Nemo 12B, Ornith 9B, Gemma4 26B A4B
- Agent Runner — adds Qwen3.6 27B and Qwen3.5 122B A10B
- Coding — adds DeepSeek V4 Flash, Qwen3.6 35B A3B, Ornith 1 35B
Models outside your plan are still available at API rates from your wallet — useful for trying a bigger model before upgrading.