The following language models are currently available on the Lehmus AI platform. The list below shows the typical use cases of each model as well as the most important supported features.
Gemma 4 31B
Use case: general-purpose, analysis, reasoning, coding
- google/gemma-4-31b-it
- public model id: jwwqblcgkizlhxjjbkcp
- enable-auto-tool-choice
- reasoning-parser=gemma4
- tool-call-parser=gemma4
- max-model-len=262144
- kv-cache-dtype=fp8
Gemma 4 26B MoE
Use case: long materials, summarization, analysis, coding
- google/gemma-4-26B-A4B-it
- public model id: wnofibsjlomanbprgbdp
- enable-auto-tool-choice
- reasoning-parser=gemma4
- tool-call-parser=gemma4
- max-model-len=262144
GPT OSS 120B
Use case: analysis, problem-solving, agents, tools
- openai/gpt-oss-120b
- public model id: azsydsttjnlbfjbgqnwd
- enable-auto-tool-choice
- reasoning-parser=openai_gptoss
- tool-call-parser=openai
- kv-cache-dtype=fp8
Qwen3.6-35B-A3B
Use case: coding, agents, reasoning, tool calls
- qwen/qwen3.6-35B-A3B
- public model id: ctraxdkjuwfrlivzrhgt
- enable-auto-tool-choice
- reasoning-parser=qwen3
- tool-call-parser=qwen3_xml
- max-model-len=131072
- max-num-batched-tokens=4096
- attention-backend=FLASHINFER
Qwen3.8-27B
Use case: general-purpose, conversation, analysis, coding
- qwen/qwen3.8-27B
- public model id: Qwen3.8-27B
- enable-auto-tool-choice
- reasoning-parser=qwen3
- tool-call-parser=qwen3_coder
- max-model-len=262144
- speculative-config.num_speculative_tokens=3
- speculative-config.method=mtp
- kv-cache-dtype=fp8
DeepSeek R1 Distill Llama 70B FP8 dynamic
Use case: coding, agents, reasoning, tool calls
- neuralmagic-deepseek-r1-distill-llama-70b-fp8-dynamic
- public model id: neuralmagic/DeepSeek-R1-Distill-Llama-70B-FP8-dynamic
- enable-auto-tool-choice
- reasoning-parser=deepseek_r1
- tool-call-parser=hermes
- max-model-len=32768
- max-num-seqs=2
- max-num-batched-tokens=8192
- trust-remote-code
- tokenizer-mode=auto
- tensor-parallel-size=1
- dtype=auto
- tokenizer=neuralmagic/DeepSeek-R1-Distill-Llama-70B-FP8-dynamic
- kv-cache-dtype=fp8
ICT Services aim to keep the model catalog as diverse as possible and to add new models whenever feasible. However, capacity is limited, so not all models can be offered. Instead, the selection is developed to support as many use cases as possible.
Is the language model you want not available?
Please first check whether you could use one of the existing language models for your purpose. You can request a new language model for the platform using the following form: requesting a new language model. After you submit a request for a new language model, our team will evaluate it based on the following criteria:
- Hardware compatibility: Assessment of VRAM requirements in relation to the available resources
- Data protection: We ensure that the model can be run safely in our environment without known side effects