AIUpdateWatch Daily Intelligence

Open models and open-source AI software

Open does not automatically mean local. This page tracks large self-hostable model releases and open-source AI software while clearly labeling the hardware or service required.

Morning edition: Friday, July 24, 2026 · Data cutoff Jul 23, 2026, 8:30 PM (America/New_York)

Looking for laptop and desktop options?

Use the separate Local Models page

The Local Models page contains consumer-practical 3B–4B starting points. The entries below may require server-class hardware or may be software that connects to a separate model.

Open Local Models →
Large open-weight model

DeepSeek V4-Pro Preview

1.6T total parameters / 49B active, 1M context

Realistic deployment
Data-center / multi-GPU deployment

Provider-published architecture information.

Why this classification: Open weights do not make a 1.6T-parameter mixture-of-experts model practical on a normal laptop or desktop.

Open original source (opens in a new tab)
Large open-weight model

DeepSeek V4-Flash Preview

284B total parameters / 13B active, 1M context

Realistic deployment
Server-class / specialist workstation

Scheduled to back common API aliases from July 24.

Why this classification: The lower active-parameter count reduces computation per token, but the full model still requires storing a very large weight set.

Open original source (opens in a new tab)
Open-source AI software

Grok Build 0.1

Open-source coding-oriented release announced July 15

Realistic deployment
Local agent harness; model accessed separately

Review license and deployment requirements before production use.

Why this classification: The open-source release is the coding agent, terminal interface and harness. The grok-build-0.1 model itself is served through the xAI API.

Open original source (opens in a new tab)

Before self-hosting or adopting open-source AI software

License

Confirm commercial use, redistribution, derivative-work and acceptable-use conditions. Software and model weights may use different licenses.

Hardware and service boundary

Identify whether you are downloading model weights, running a local agent that calls a remote API, or deploying a multi-GPU model server.

Operations

Plan patching, model provenance, security review, evaluation, monitoring, access control and rollback.