Skip to content

Available Models

Models are organized by cluster and marked with the following capabilities:

  • B - Batch Processing Enabled
  • T - Tool Calling Enabled
  • R - Reasoning Enabled
  • H - Always Hot Model

The table reports each endpoint's cluster, type, organization, API protocols, context window (in tokens), supported input modalities, streaming support, reasoning, and tool calling. A — indicates the value is not published for that endpoint.

Cluster Type Organization Model API Protocols Context Window Input Modalities Streaming Reasoning Tool Calling
Sophia Chat Arcee AI arcee-ai/Trinity-Large-Thinking-W4A16RT chat_completions 131072 text ✓ ✓ ✓
Sophia Chat Argonne argonne/AuroraGPT-DPO-UFB-0225B chat_completions — text ✓ — —
Sophia Chat Argonne argonne/AuroraGPT-IT-v4-0125B chat_completions — text ✓ — —
Sophia Chat Argonne argonne/AuroraGPT-KTO-UFB-0325B chat_completions — text ✓ — —
Sophia Chat Argonne argonne/AuroraGPT-Tulu3-SFT-0125B chat_completions — text ✓ — —
Sophia Chat AstroMLab AstroMLab/AstroSage-70B-20251009T chat_completions — text ✓ — ✓
Sophia Chat Google google/gemma-3-27b-itBT chat_completions 32768 text, image ✓ — ✓
Sophia Chat Google google/gemma-4-26B-A4B-itRT chat_completions 262144 text ✓ ✓ ✓
Sophia Chat Google google/gemma-4-31B-itRTH chat_completions 128000 text, image, video ✓ ✓ ✓
Sophia Chat Google google/gemma-4-E4B-itRTH chat_completions — text, image, video ✓ ✓ ✓
Sophia Chat Meta meta-llama/Llama-3.3-70B-InstructBT chat_completions 32768 text ✓ — ✓
Sophia Chat Meta meta-llama/Llama-4-Maverick-17B-128E-InstructT chat_completions 16384 text, image ✓ — ✓
Sophia Chat Meta meta-llama/Llama-4-Scout-17B-16E-InstructBT chat_completions 32768 text, image ✓ — ✓
Sophia Chat Meta meta-llama/Meta-Llama-3.1-405B-InstructBT chat_completions 24576 text ✓ — ✓
Sophia Chat Meta meta-llama/Meta-Llama-3.1-70B-InstructBTH chat_completions 16384 text ✓ — ✓
Sophia Chat Meta meta-llama/Meta-Llama-3.1-8B-InstructBTH chat_completions 32768 text ✓ — ✓
Sophia Chat Mistral AI mistralai/Devstral-2-123B-Instruct-2512T chat_completions 16384 text ✓ — ✓
Sophia Chat Mistral AI mistralai/Mistral-Large-Instruct-2407 chat_completions 16384 text ✓ — —
Sophia Chat Mistral AI mistralai/Mixtral-8x22B-Instruct-v0.1 chat_completions 16384 text ✓ — —
Sophia Chat NVIDIA mgoin/Nemotron-4-340B-Instruct-hf chat_completions — text ✓ — —
Sophia Chat NVIDIA nvidia/nemotron-3-super-120bRT chat_completions 262144 text ✓ ✓ ✓
Sophia Chat OpenAI openai/gpt-oss-120bBRTH chat_completions 65536 text ✓ ✓ ✓
Sophia Chat OpenAI openai/gpt-oss-20bBRTH chat_completions — text ✓ ✓ ✓
Sophia Vision Meta meta-llama/Llama-3.2-90B-Vision-InstructT chat_completions 8192 text, image ✓ — ✓
Sophia Embedding Google google/embeddinggemma-300m — — — — — —
Sophia Embedding Mistral AI mistralai/Mistral-7B-Instruct-v0.3-embed — — — — — —
Sophia Embedding Salesforce Salesforce/SFR-Embedding-Mistral — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-300M-aminoacid — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-300M-codon — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-300M-contrastive — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-300M-contrastive-2026-08 — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-300M-joint — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-600M-aminoacid — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-600M-codon — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-600M-contrastive — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-600M-contrastive-2026-08 — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-600M-joint — — — — — —
Sophia Genomics genslm genslm-test/genslm-esmc-6B-contrastive-2026-08 — — — — — —
Sophia Segmentation Meta dinov3 — — — — — —
Sophia Segmentation Meta facebook/sam3 — — — — — —
Sophia Triton — amsc-d3 KServe gRPC — — — — —
Metis Chat Google gemma-4-31B-itRTH chat_completions, responses 262144 text, image, video ✓ ✓ ✓
Metis Chat Mistral AI Mistral-Large-3-675B-Instruct-2512TH chat_completions 32768 text ✓ — ✓
Metis Chat OpenAI gpt-oss-120bRTH chat_completions, responses 131072 text ✓ ✓ ✓
Minerva Chat Meta muse-spark-bf16RTH chat_completions, responses, messages 1048576 text, image ✓ ✓ ✓
Minerva Chat Mistral AI mistral-large-4-nvfp4RTH chat_completions, responses, messages 1048576 text, image ✓ ✓, separate (low, medium, high; default high) ✓
Minerva Chat Reflection AI beam-1.0-bf16RTH chat_completions, responses, messages 1048576 text ✓ ✓ ✓
Minerva Chat Thinking Machines inkling-bf16RTH chat_completions, responses, messages 1048576 text, image ✓ ✓ ✓
Tara Chat NVIDIA nemotron-3-ultraRT chat_completions 1048576 text ✓ ✓ ✓
Tara Chat NVIDIA nemotron-35-super-vlRT chat_completions, responses, messages 1048576 text, image, video ✓ ✓ ✓
Tara Chat OpenAI openai/gpt-oss-20bRT chat_completions, messages, responses 65536 text ✓ ✓ ✓
Tara Chat Thinking Machines inkling-bf16RT chat_completions 1048576 text, image, audio ✓ ✓ ✓

For Metis, endpoint and model status is on the Metis status page. See SambaNova's OpenAI compatible API documentation for API details.

Metis Tool Calling

Tool calling is advertised by the API, but SambaNova's sanitization of tool calls has a known issue. See the Metis Tool Calling warning under Shims/Proxies.

Promptable Image Segmentation Models

The SAM 3 and DINOv3 models are deployed on Sophia. Install the alcf-ai package for the command line and Python toolkit. See alcf-ai CLI and SDK for worked examples, including batch and DINOv3 segmentation.

If this is your first time using alcf-ai, run alcf-tokens login inference first. Without a valid token, alcf-ai exits with an authentication error naming that command.

Model Serving Configuration

When available, model serving configuration details can be viewed for each cluster.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
#!/bin/bash

# Get your access token
access_token=$(alcf-tokens get-token inference)

# Check serving configuration for all Sophia models
curl -X GET "https://inference-api.alcf.anl.gov/resource_server/sophia/models" \
    -H "Authorization: Bearer ${access_token}"

# Check serving configuration for a specific model (e.g., openai/gpt-oss-120b)
curl -X GET "https://inference-api.alcf.anl.gov/resource_server/sophia/models?model_id=openai/gpt-oss-120b" \
    -H "Authorization: Bearer ${access_token}"

# Check serving configuration for another cluster (e.g., Tara)
curl -X GET "https://inference-api.alcf.anl.gov/resource_server/tara/models" \
    -H "Authorization: Bearer ${access_token}"

# Check serving configuration for Minerva
curl --fail-with-body "https://inference-api.alcf.anl.gov/resource_server/minerva/models" \
    -H "Authorization: Bearer ${access_token}"

Model Capabilities

Each model includes a versioned capabilities object describing what the deployment supports. The current schema is version 1:

Field Description
schema_version Version of the capability object.
api_protocols API protocols the model accepts.
context_window_tokens Maximum context window, in tokens.
input_modalities Accepted input types.
streaming Whether streaming responses are supported.
reasoning.supported Whether the model supports reasoning.
reasoning.effort_levels Accepted reasoning effort values.
reasoning.default_effort Effort applied when a request omits one.
reasoning.separate_output Whether reasoning is returned separately from final content; the field or block type depends on the API protocol.
tool_calling.supported Whether the model supports tool calling.

Optional fields are omitted when they do not apply, so clients should fall back to conservative defaults. Models served by vLLM also report max_model_len, max_num_seqs, tool_call_parser, and enable_auto_tool_choice alongside capabilities.

Context limits include input and generated output; leave room for reasoning and the final answer. Accepted reasoning efforts are model-specific, not a universal set. See API reasoning fields and agent configuration for how clients use these capabilities.

Possible Values

Field Values
schema_version 1
api_protocols chat_completions, responses, messages
context_window_tokens Any positive integer
input_modalities text, image, video
streaming true, false
reasoning.supported true, false
reasoning.effort_levels none, minimal, low, medium, high, xhigh, max
reasoning.default_effort One of the model's reasoning.effort_levels
reasoning.separate_output true, false
tool_calling.supported true, false

alcf-ai ls-models <cluster> returns the same objects and uses them to configure agent harnesses. See alcf-ai CLI and SDK.

Want to add a model?

To request a new model, please contact ALCF Support.