Supported models

Models on sovereign hardware.

Argyll currently runs MiniMax M2.7 on our dedicated UK infrastructure. The models below are supported for customers who require custom configurations, tailored deployments, or specific routing requirements.

Talk to sales

424.8 tok/s on M2.7

Up to 6× faster than GPU-based providers. SambaNova RDU architecture delivers industry-leading throughput with the lowest time-to-first-token.

MiniMax-M2.7

MiniMax

Primary ModelText

Context lengths

8K–32K (BS 2, 4, 6, 8) · 160K (BS 2) · 192K (BS 2)

Capabilities

Function calling · Structured output

Coding agentSWE-Pro 56.2%GDPval-AA ELO 1,494

Model cardChat completions

gpt-oss-120b

OpenAI

FeaturedText

Context lengths

8K–32K (BS 2, 4, 6, 8) · 64K (BS 2, 4) · 128K (BS 2)

Capabilities

Reasoning · Function calling · JSON mode · Logit masking

Main / planner agentTool-calling agent

Model cardChat completions

gpt-oss-20b

OpenAI

Text

Context lengths

8K–32K (BS 2, 4, 6, 8) · 64K (BS 2, 4) · 128K (BS 2)

Capabilities

Function calling · JSON mode

Main / planner agentTool-calling agentReasoning

Model cardChat completions

MiniMax-M2.5

MiniMax

Text

Context lengths

8K–32K (BS 2, 4, 6, 8) · 160K (BS 2) · 192K (BS 2)

Capabilities

Function calling · Structured output

Coding agent

Model cardChat completions

DeepSeek-R1-0528

DeepSeek

ReasoningText

Context lengths

4K (BS 4) · 8K (BS 1) · 16K (BS 1) · 32K (BS 1)

Capabilities

Function calling · JSON mode

Complex reasoning

Model cardChat completions

DeepSeek-V3.2

DeepSeek

Text

Context lengths

8K (BS 1, 4) · 16K (BS 1) · 32K (BS 1) · 128K (BS 1)

Capabilities

Optional thinking mode · Function calling · JSON mode

Main / planner agentTool-calling agent

Model cardChat completions

DeepSeek-V3.1

DeepSeek

ReasoningText

Context lengths

4K (BS 4) · 8K (BS 1, 4) · 16K (BS 1) · 32K (BS 1)

Capabilities

Function calling · JSON mode

Main / planner agentTool-calling agent

Model cardChat completions

DeepSeek-V3-0324

DeepSeek

Text

Context lengths

4K (BS 4) · 8K (BS 1, 4) · 16K (BS 1) · 32K (BS 1)

Capabilities

Function calling · JSON mode

Main / planner agentTool-calling agent

Model cardChat completions

DeepSeek-R1-Distill-Llama-70B

DeepSeek

ReasoningText

Context lengths

4K–128K · Up to BS 32 at shorter contexts

Capabilities

Speculative decoding · Custom checkpoints

Complex reasoning

Model cardChat completions

Meta-Llama-3.3-70B-Instruct

Meta

Text

Context lengths

4K–128K · Up to BS 32 at shorter contexts

Capabilities

Function calling · JSON mode · Speculative decoding · Custom checkpoints

Task agentTool-calling agentText to SQL / Cipher

Model cardChat completions

Meta-Llama-3.1-8B-Instruct

Meta

Text

Context lengths

4K (BS up to 128) · 8K (BS up to 64) · 16K (BS up to 8)

Capabilities

Function calling · JSON mode · Custom checkpoints

Gateway agentValidation agent

Model cardChat completions

Meta-Llama-3.1-405B-Instruct

Meta

Text

Context lengths

4K (BS 1, 2, 4) · 8K (BS 1) · 16K (BS 1)

Capabilities

Function calling · JSON mode · Speculative decoding

Task agentTool-calling agentCode generation

Model cardChat completions

Llama-4-Maverick-17B-128E-Instruct

Meta

ImageText

Context lengths

8K–128K (BS 1)

Capabilities

Function calling · JSON mode

Image understandingTask agentTool-calling agent

Model cardChat completions

Mistral-Large-3-675B-Instruct-2512

Mistral AI · Preview

Text

Context lengths

8K (BS 1)

Capabilities

Function calling

MultilingualTask agentTool-calling agent

Model cardChat completions

Qwen3-235B-A22B-Instruct-2507

Alibaba Cloud

ReasoningText

Context lengths

32K (BS 2, 4, 6, 8) · 128K (BS 2)

Capabilities

Non-thinking mode · Function calling · JSON mode

Agentic plannerMultilingual

Model cardChat completions

Qwen3-32B

Alibaba Cloud

ReasoningText

Context lengths

8K (BS 1, 4) · 16K (BS 1) · 32K (BS 1, 2)

Capabilities

Thinking mode · Tool use · Multilingual

Task agentMultilingual

Model cardChat completions

gemma-3-27b-it

Google

Text

Context lengths

4K–128K (BS 2, 4, 6, 8)

Capabilities

Image understanding · JSON mode

Image understandingTask agent

Model cardChat completions

gemma-3-12b-it

Google

ImageText

Context lengths

128K (BS 2, 4, 6, 8)

Capabilities

Image understanding · JSON mode

Image understandingTask agent

Model cardChat completions

gemma-4-31B-it

Google · Preview

Text

Context lengths

128K (BS 2, 4, 6, 8)

Capabilities

Function calling · JSON mode

Task agent

Model cardChat completions

Whisper-Large-v3

OpenAI

Audio

Context lengths

4K (BS 1, 16, 32)

Capabilities

Audio transcription · Speech translation · Multilingual ASR

Speech recognition (ASR)Audio transcription

Model cardTranslation · Transcription

E5-Mistral-7B-Instruct

intfloat

Embedding

Context lengths

4K (BS 1, 4, 8, 16, 32)

Capabilities

Text embeddings · Semantic search · Retrieval

Vector storage & retrieval (RAG)

Model cardEmbeddings

All models served from UK-sovereign infrastructure via an OpenAI-compatible API.
Swap your base URL. Keep your code.

Start building Talk to sales