Two new models are now available in the Ollama library: Mistral Small 3.1 and DeepCoder-14B-Preview. The Mistral Small 3.1 model shows significant improvements in text performance, understanding of multimodal data, and has an expanded context window up to 128,000 tokens. It surpasses the Gemma 3 and GPT-4o Mini models while maintaining a token output speed of 150 tokens per second. The model is released under the Apache 2.0 license and operates on a single RTX 4090 or Mac with 32 GB of RAM.
The DeepCoder-14B-Preview model stands out for its high accuracy in logical code analysis, achieving a Pass@1 rate of 60.6% on LiveCodeBench. This exceeds the performance of previous versions by 8%. The model was trained using distributed RL based on Deepseek-R1-Distilled-Qwen-14B and uses just 14 billion parameters.
Users can install these models in AI chat-bots, including the option to rent a server hourly for testing on the Ollama platform.