Liquid AI releases open-weight LFM2.5-24B-A2B reasoning model for local agents
Article timeline
The articles in this story. Times are ET.
OpenRouter@OpenRouterOfficialOpenRouter lists Liquid's LFM 2.5 24B A2B model for testingAIOpenRouter links to a page where users can try the Liquid LFM 2.5 24B A2B model. The post gives no benchmark scores, pricing, or context length.
- OpenRouter@OpenRouterOfficialLFM2.5-24B-A2B from Liquid AI is now on OpenRouter
AIOpenRouter now offers LFM2.5-24B-A2B from Liquid AI, a 24B-parameter mixture-of-experts model with 2B active parameters per token. The model reasons before answering and is built for agents and tool use.
Liquid AI@liquidaiOfficialLiquid AI's LFM2.5-24B-A2B gets Day-0 support in SGLangAISGLang adds Day-0 support for Liquid AI's LFM2.5-24B-A2B, a 24B MoE reasoning model with a 131K context window. The model scores 48.09 on LiveClaw and 27.69 on AutomationBench, versus 38.19 and 8.72 for Qwen3.6-35B-A3B. It reaches up to 21K output tokens per second on a single NVIDIA B300.
SGLang@sgl_projectOfficialSGLang adds Day-0 support for Liquid AI's LFM2.5-24B-A2B modelAISGLang announces Day-0 support for LFM2.5-24B-A2B, a 24B-parameter mixture-of-experts reasoning model from Liquid AI with 2B active parameters and a 131K context window. The post reports 48.09 versus 38.19 on LiveClaw and 27.69 versus 8.72 on AutomationBench against Qwen3.6-35B-A3B, and up to 21K output tokens per second on a single NVIDIA B300. Liquid AI's post says the open base and post-trained weights are available on Hugging Face.
- Liquid AI@liquidaiOfficialLiquid AI releases LFM2.5-24B-A2B model with open weights and API access
AILiquid AI releases LFM2.5-24B-A2B, a 24B-parameter model with 2B active parameters, with weights on Hugging Face and an API at console.liquid.ai. The post links to a blog, documentation, and a page for custom tuning jobs.
- Liquid AI@liquidaiOfficialLiquid AI says post-training runs reinforcement learning inside agent harnesses
AILiquid AI says its post-training includes reinforcement learning inside agent harnesses such as OpenClaw, Hermes Agent by Nous Research, pi, and opencode. The model learns from tool calls and multi-turn tasks in those environments.
- Liquid AI@liquidaiOfficialLiquid AI runs open-source Hermes Agent on LFM2.5-24B-A2B locally
AILiquid AI demonstrates the open-source desktop agent Hermes Agent running on its LFM2.5-24B-A2B model on a single laptop. The demo uses local model inference with no cloud model API calls or API keys.
- Liquid AI@liquidaiOfficialPickLiquid AI releases LFM2.5-24B-A2B, its largest LFM2.5 model for local agent tasks
AILiquid AI releases LFM2.5-24B-A2B, the largest member of its LFM2.5 family, as an efficient local model for laptops and PCs running multi-step tasks in popular agent harnesses. The model has 24B total parameters, 2B active per token, a 131K-token context window, a 128K-token vocabulary and 33T pretraining tokens. Liquid AI says it performs competitively with Qwen3.6-35B-A3B on agentic and tool-use tasks, and open base and post-trained weights are available on Hugging Face.
- Liquid AI@liquidaiOfficialLiquid AI releases LFM2.5-24B-A2B with day-one support across inference runtimes
AILiquid AI says its LFM2.5-24B-A2B model supports llama.cpp and GGUF, MLX for Apple Silicon, vLLM, SGLang, and ONNX on day one. The company reports 184 tokens/s decoding on an Apple M5 Max. Users can download the weights and fine-tune and deploy the model on their own hardware.
- Hugging Face BlogOfficialPickLiquid AI releases LFM2.5-24B-A2B, a 2B-active reasoning model for on-device agents
AILiquid AI releases LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters per token, as open weights on Hugging Face. On its own benchmarks it scores 48.09 on LiveClaw versus 38.19 for Qwen3.6-35B-A3B and leads on five tool-use and agentic tests, but trails larger models on knowledge-heavy and coding benchmarks such as GPQA Diamond and SWE-Bench Verified (OpenCode). It reports 184 tokens per second on an Apple M5 Max and up to 21K output tokens per second on a single NVIDIA B300.
- Liquid AI BlogOfficialPickLiquid AI releases LFM2.5-24B-A2B, a 24B on-device agent model
AILiquid AI releases LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters built for on-device agents. The company says it beats Qwen3.6-35B-A3B on five of eight agentic and tool-use benchmarks, scoring 48.09 on LiveClaw versus 38.19 and 27.69 on AutomationBench versus 8.72. It reports 184 tokens/s on an Apple M5 Max and up to 21K output tokens/s on a single NVIDIA B300 GPU, with open weights on Hugging Face.
- OpenRouter · New modelsBlogLiquid AI releases LFM2.5-24B-A2B, a 24B mixture-of-experts reasoning model
AILiquid AI's LFM2.5-24B-A2B is a mixture-of-experts reasoning model with 24B total and 2.45B active parameters. The model always reasons before answering and is built for agents that call tools. The feed supplied only an excerpt, so further details are not available from this source.
- Liquid AI · new models on Hugging FaceOfficialPickLiquid AI releases LFM2.5-24B-A2B, a 24B reasoning model for on-device agents
AILiquid AI releases LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters, aimed at fast, private on-device agents. The card reports 48.09 on LiveClaw against 38.19 for Qwen3.6-35B-A3B and 184 tokens/s on an Apple M5 Max. Weights are available in Transformers, FP8, GGUF, ONNX, and MLX-4bit formats, with a 131,072-token context window.