Anthropic
Amazon Investment
Amazon has invested another $4 billion on Anthropic, bringing their total investment to $8 billion.
Amazon
Foundation Models
Amazon released a family of foundation models. These 4 models are named Nova Micro, Nova Lite, Nova Pro and Nova Premier and have different capabilities. Amazon published performance results for the models. The family also includes Nova Canvas, a text to image generation model and Nova Reel, a text to video generation model.
Google
Veo and Imagen
Google had previously announced the image-to-video model Veo and text-to-image model Imagen last May 2024. They now released these two models to Vertex AI (platform to train and use AI models) customers in preview mode.
Gemini 2.0
Google released a new version of their LLM, Gemini 2.0. The novelty in this version is the introduction of AI Agents.
Veo 2.0
Google released the new version of their image-to-video model Veo with the capability to generate 4K videos. The output is really impressive, with finer control over the generated video.
OpenAI
12 Days of OpenAI
OpenAI has announced that they will have 12 announcements in December in the spirit of 12 days of Christmas. Some of the announcements were more like repetitions of some features previously released.
Day 1: o1 and ChatGPT Pro
OpenAI has now made the o1 model available for paid subscriptions, improving the model from its preview version.
OpenAI has also introduced ChatGPT Pro, a $200 ChatGPT subscription, which includes unlimited access to all OpenAI models. OpenAI also announced that o1 is now performing much better in math and higher reasoning, although I’ve read some reviews that this does not always seem to be the case.
Day 3: Sora
OpenAI has now opened the text-to-video generation model Sora to general public use, although there is a glitch with account creation at the moment. OpenAI also declared that a faster version of the model dubbed Sora Turbo is released.
Day 4: Canvas
OpenAI had introduced the Canvas feature that was a kind of workplace where the prompter and ChatGPT could collaborate. It is now available to all users, although it may take a while to deploy updated versions of desktop or mobile versions. Currently only the web version seems to have it.
Day 8: SearchGPT
OpenAI has now opened the SearchGPT functionality (which can search the web with prompts) to all users. Currently it seems to be available on the web versions only, but should soon appear on native applications as well.
Day 12: o3 and o3 Mini
OpenAI announced the next frontier model, called o3 (since o2 is a well-known cellular plan and company name). They reported benchmark results that seemed to mean that this is an incredibly strong frontier model. o3 mini is a performance-and-cost-focused model and its performance was demonstrated using code generation. Both models are open to security and safety researchers to test and the public launch is expected end of January for o3 mini and shortly after for o3.
Both models have three reasoning modes (low, medium, high) which affect the response time and cost.