# Updates · Good AI Labs

> New experiments, reports and openings at the lab.

URL: https://www.goodailabs.com/updates/
Date: 2026-10-05

## Entries

- 2026-10-05: Reflex-1: the latest public weights, reserved evaluation results and CPU measurements are now on the model page. 421M parameters; 2.17 GiB peak process memory in the recorded CPU replay.
- 2026-10-05: Samsara-VLA and Samsara-VLA Tiny: watch eight recorded robot tasks across four LIBERO suites. Compare the 217M and 137M policies, their success rates and local inference options.
- 2026-10-02: reflex-1 is available on Hugging Face with public FP32 weights, tokenizers and Transformers inference code. The model page includes dated game and browser recordings, evaluation conditions and resource measurements.
- 2026-10-02: Samsara VLA: published all forty task counts and a CSV of 10,000 recorded trials across five checkpoints, with architecture and training-resource details.
- 2026-10-02: re1: explore the research preview through recorded examples of video recall and captured environments.
- 2026-10-02: Colony: persistent personal agents, scheduled work, voice, Mac computer use and small apps. Read the current technical overview.
- 2026-10-01: Announced reflex-1 and Samsara VLA. Detailed model pages include latency, accuracy and resource measurements.
- 2026-09-29: reflex-1 H200 test: 27.5 ms median and 30.5 ms p95 across 240 recorded requests, using FP32 and loopback HTTP.
- 2026-09-24: reflex-1 M4 CPU test: 345.2 ms median for Banking77 intent selection, using FP32 and one thread.
- 2026-09-17: Samsara VLA completed 1,897 of 2,000 LIBERO episodes. The selected checkpoint scored 94.85% overall and 99.6% on Object.
- 2026-08-19: Open roles in robot learning, 3D vision, agent systems and deployment engineering.
- 2026-08-18: Journal: local inference and hardware requirements.
- 2026-08-12: Four robot-grading protocols, with 400-episode results across three historical checkpoints.
- 2026-08-05: Added a plain-text version of each page at llms.txt.
- 2026-07-14: Journal: development evaluations and reserved confirmation sets.
- 2026-05-06: Journal: deploying models inside a local application.
- 2026-02-11: Journal: the lab's reasons for working on local models.
- 2025-03-11: Research notes on model weights and adaptation.

Announcements and engineering notes are in the [journal](/blog/).

Contact: research@goodailabs.com