News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM

API and app only

Used through an API or an app, weights kept private: GPT-4, Claude, Gemini. 150 models on EveryLLM so far.

API-only models are offered as a service. Developers send text, images or audio to the maker's servers through an application programming interface (API) and pay by the token; the public uses the same models in apps such as ChatGPT, Claude and Gemini. The weights never leave the company, which keeps control over updates, safety filters and pricing, and protects what it regards as its most valuable asset.

How it works

Each call is billed by the tokens going in and coming out, with lower rates for smaller models and for prompts the provider has cached. Providers publish model cards and system cards describing testing and limits, and retire old versions on a schedule. Cloud platforms resell many of these models: Microsoft's Azure offers OpenAI's, while Amazon Bedrock and Google Cloud's Vertex AI offer Anthropic's Claude. Large customers can sign agreements that keep their data out of training.

Trade-offs

Closed models have usually led on the hardest benchmarks, and a managed service spares users the cost of running large hardware. In return, users depend on a vendor's uptime, terms and decisions to change or retire a model, and outsiders cannot inspect the weights. Safety researchers and regulators get access through agreements: the UK and US AI safety institutes tested several frontier models before their release from 2024.

Famous API only models

Still to be written: GPT-4o.

Further reading

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.