Available models¶
alcf-proxy forwards to whatever models ALCF serves. This page mirrors ALCF's
official catalog; run alcf-proxy models
for the live, authoritative list (ids differ per cluster).
Recommended: ⭐ openai/gpt-oss-120b
openai/gpt-oss-120b is the recommended default for coding CLIs — it's a large reasoning model,
always hot (low latency), supports tool calling, reasoning, and batch, and is
available on both Sophia (openai/gpt-oss-120b) and Metis (gpt-oss-120b). It is alcf-proxy's
built-in default and failover chain.
Feature legend¶
| Flag | Meaning |
|---|---|
| B | Batch processing enabled |
| T | Tool calling enabled |
| R | Reasoning enabled |
| H | Always hot (kept loaded → low latency) |
Sophia cluster (sophia/vllm)¶
Base URL: …/resource_server/sophia/vllm/v1 · Endpoints: chat/completions, responses, messages,
completions, embeddings, batches.
Chat / language models¶
| Model | B | T | R | H |
|---|---|---|---|---|
| ⭐ openai/gpt-oss-120b | ✅ | ✅ | ✅ | ✅ |
| openai/gpt-oss-20b | ✅ | ✅ | ✅ | ✅ |
| meta-llama/Meta-Llama-3.1-8B-Instruct | ✅ | ✅ | ✅ | |
| meta-llama/Meta-Llama-3.1-70B-Instruct | ✅ | ✅ | ✅ | |
| meta-llama/Meta-Llama-3.1-405B-Instruct | ✅ | ✅ | ||
| meta-llama/Llama-3.3-70B-Instruct | ✅ | ✅ | ||
| meta-llama/Llama-4-Scout-17B-16E-Instruct | ✅ | ✅ | ✅ | |
| meta-llama/Llama-4-Maverick-17B-128E-Instruct | ✅ | |||
| mistralai/Mistral-Large-Instruct-2407 | ||||
| mistralai/Mixtral-8x22B-Instruct-v0.1 | ||||
| mistralai/Devstral-2-123B-Instruct-2512 | ||||
| argonne/AuroraGPT-IT-v4-0125 | ✅ | |||
| argonne/AuroraGPT-Tulu3-SFT-0125 | ✅ | |||
| argonne/AuroraGPT-DPO-UFB-0225 | ✅ | |||
| argonne/AuroraGPT-KTO-UFB-0325 | ✅ | |||
| google/gemma-3-27b-it | ✅ | ✅ | ✅ | |
| google/gemma-4-26B-A4B-it | ✅ | ✅ | ||
| google/gemma-4-31B-it | ✅ | ✅ | ✅ | |
| google/gemma-4-E4B-it | ✅ | ✅ | ✅ | |
| arcee-ai/Trinity-Large-Thinking-W4A16 | ✅ | ✅ | ||
| nvidia/nemotron-3-super-120b | ✅ | ✅ | ||
| mgoin/Nemotron-4-340B-Instruct-hf | ||||
| AstroMLab/AstroSage-70B-20251009 |
Vision language models¶
| Model | B | T | R | H |
|---|---|---|---|---|
| meta-llama/Llama-3.2-90B-Vision-Instruct |
Embedding models¶
mistralai/Mistral-7B-Instruct-v0.3-embed, Salesforce/SFR-Embedding-Mistral, and the
genslm-test/genslm-esmc-* family (300M/600M, aminoacid/codon variants). Use the /v1/embeddings
endpoint directly against ALCF for these.
Image segmentation¶
facebook/sam3 (promptable segmentation) — see ALCF's alcf-ai toolkit; not exposed through
alcf-proxy's chat/responses/messages routes.
Metis cluster (metis/api, SambaNova)¶
Base URL: …/resource_server/metis/api/v1 · Endpoints: chat/completions, responses.
| Model | B | T | R | H |
|---|---|---|---|---|
| ⭐ gpt-oss-120b | ✅ | |||
| Llama-4-Maverick-17B-128E-Instruct | ✅ | |||
| Mistral-Large-3-675B-Instruct-2512 | ✅ | |||
| gemma-4-31B-it | ✅ |
Metis limitations
Batch processing and tool calling are not supported on Metis. Model ids on Metis have no
openai/ prefix (e.g. gpt-oss-120b, not openai/gpt-oss-120b).
Notes¶
- Coming soon (per ALCF): Cerebras CS-3 and GH200 Nvidia clusters.
- The catalog changes over time —
alcf-proxy modelsand ALCF's official docs are the source of truth. - Feature flags (B/T/R/H) are ALCF's; alcf-proxy passes tool-calling and reasoning through, so a model marked T/R works for agentic coding CLIs.