When will OpenAI or Google/DeepMind next release a frontier language model?
A frontier model is a new model that surpasses the developer's previous most powerful model in general text and coding capabilities. It does not necessarily have to be #1 across all companies.
I'll defer to the company's characterization of the model to determine whether it is better than the company's previous models, unless there's very strong reason for doubt.
"New model" roughly means a fresh training run and will be determined using naming scheme or other statements from the developer. An incremental update to an existing model, like Gemini 1.5 Pro 0801, does not count. But Gemini 1.5 Ultra would count.
1,000
3.00Anyone object to resolving today based on o1-preview? (o1-preview being the most capable o1 model available today). There may be mild regressions in some areas vs 4o but clearly a big advancement overall. And some features like reading PDFs and images haven't been turned on but that's not an issue with the model itself (hence not a reason to think it doesn't define the frontier)