ACS Research · Base model endpoint

Base models, served for research.

Direct, token-level access to large language base models, over an OpenAI compatible API. Free for researchers.

We aim to support research in AI safety and interpretability, understanding LLM psychology and personas, cyborgism-style human-AI collaboration, and more.

Apply for access Login

What you get

An endpoint built for research on the model itself, not just its outputs.

True base models

Raw /v1/completions — no chat template, no RLHF, no system prompt. The model as it came out of pretraining.

Token-level access

Top-k logprobs on generated tokens and prompt_logprobs on the input — for likelihood and interpretability work. Plus echo and SSE streaming.

OpenAI compatible

Point any OpenAI client at one base URL. Full sampling controls — temperature, top_p, top_k, min_p, penalties, and seed.

Browser workbench

Prompt the models straight from the browser to get a feel for them before you (ask your friendly LLMs to) write a line of code.

Models

A small set is live today; beyond that we'll figure out with our beta tester community what to add and on what schedule. See the models page for the current list and live capabilities.

llama-8b always-on starter
llama-405b
trinity-truebase

Candidates we're weighing up (not yet served):

Getting access

We're in an open beta. Apply with the signup form — a quick note on who you are and how you'd use it — and we'll review and approve individually. We can't promise access or a timeline.

Apply

Fill in the signup form — who you are and what you'd like to work on.

Get approved

We review each application and email you when it's approved.

Create a key

Generate an API key and start querying.

Need base models for your research?

Apply for access with the signup form, or get in touch if you have questions about what's available.

Apply for access infra@acsresearch.org