0.9

• Track token usage using new capability introduced in LLM 0.19. #15

0.9a0

• Track token usage. #15

0.8

• Provide async versions of the Mistral models, compatible with LLM 0.18. #13
• Added a pixtral-large alias for the new Pixtral Large model. #14

  llm -m pixtral-large describe -a https://static.simonwillison.net/static/2024/pelicans.jpg

0.7

• Support for sending images to the pixtral-12b vision model using the new llm 'prompt' -a image.jpeg option. #12

0.6

• Fixed an issue where an invalid API key caused llm models to fail with an error. #10
• Added new aliases for Ministral 3B and 8B: llm -m ministral-3b hi and llm -m ministral-8b hi. #11

0.5

• New alias llm -m mistral-nemo "prompt goes here" for Mistral's NeMo model. Thanks, Thaddée Tyl #9
• The new Mistral Large 2 model is already supported by the llm -m mistral-large alias, since that alias uses the latest release in the Large series.

0.4

• Documentation for the llm mistral refresh command, which can be used to refresh the list of available Mistral API models.
• New default aliases: llm -m codestral for the latest release of Codestral and llm -m codestral-mamba for the latest release of the new Codestral Mamba. #8

0.3.1

• No longer raises an error if you run llm models without first setting a mistral API key. #6
• Mixtral 8x22b is now available as llm -m mistral/open-mixtral-8x22b 'say hello'. New installations will get this model automatically - if you do not see the model in the llm models list you should run llm mixtral refresh to update your local cache of available models. #7

0.3

• Support for the new Mistral Large model - llm -m mistral-large "prompt goes here". #5
• All Mistral API models are now supported automatically - LLM fetches a list of models from their API the first time the plugin is installed, and that list can be refreshed at any time using the new llm mistral refresh command.
• When using the Python API a model key can now be set using model.key = '...' - thanks, Alexandre Bulté. #4

0.2

• Mistral LLM models now support options: -o temperature 0.7, -o top_p 0.1, -o max_tokens 20, -o safe_mode 1, -o random_seed 12. #2
• Support for the Mistral embeddings model, available via llm embed -m mistral-embed -c 'text goes here'. #3
• The --no-stream option now uses the non-streaming Mistral API.