Daily briefing
October 16, 2026
2 stories · about 2 minutes
Tesla reveals the new Roadster, and the top version can lift off the ground
Nine years after the first reveal, Tesla showed the second-generation Roadster in Texas with SpaceX cold-gas thrusters, three trims, $50,000 reservations and production planned for 2028.
- Tesla revealed the second-generation Roadster at SpaceX's test site in McGregor, Texas, where the car hung upside down on a ramp and used thrusters to lift itself into the air.
- It comes in three trims, Performance, Thrust and Rocket, and the single-seat Rocket has 10 cold-gas thrusters fed by compressed air.
- Tesla is taking reservations of $50,000 for Performance and $250,000 for Thrust, while Rocket is invite only and no final price has been named.
- Elon Musk says production starts in 2028, and aviation experts told Wired that flying it legally would need FAA approval and a pilot's license.
Why it matters: Tesla revealed its second-generation Roadster in Texas, and its top Rocket trim used 10 SpaceX-derived cold-gas thrusters to lift the car off the ground. Tesla is taking $50,000 reservations for the base car and $250,000 for the Thrust version, but has not named a final price. The car arrives nine years after its first reveal. Elon Musk says production starts in 2028, and experts say flying it would need FAA approval.
Read the full story →Microsoft finishes training MAI-Base-2, a 2.8-trillion-parameter AI model
Microsoft AI says its new in-house base model beats every open-weight model it tested and trained far more efficiently than DeepSeek's best. Nobody outside can use it yet.
- Microsoft AI finished pre-training MAI-Base-2, a mixture-of-experts model with 2.8 trillion total parameters and 80 billion active ones.
- It trained for 38 days on 16,000 Nvidia GB200 GPUs in August and September 2026, on a 50-trillion-token mix that included images from the start.
- Microsoft says it beats every open-weight model it tested and is 4 to 12 times more compute-efficient to pre-train than DeepSeek-V4-Pro-Base.
- The model is now in post-training, and Microsoft plans to license its weights to trusted partners through Foundry.
Why it matters: Microsoft AI finished pre-training MAI-Base-2, an in-house model with 2.8 trillion total parameters, of which 80 billion work on each word. It trained for 38 days on 16,000 Nvidia GB200 GPUs using 50 trillion tokens of text and images. Microsoft says it beats every open-weight model it tested and was 4 to 12 times more efficient to train than DeepSeek-V4-Pro-Base. The model now goes through post-training before Microsoft licenses it to partners through Foundry.
Read the full story →