Grok 4.5, based on our 1.5T V9 foundation model, with Cursor data added in supplemental training, is now in private beta at SpaceX & Tesla. Early evals show performance close to, perhaps exceeding Opus.
RL is continuing to significantly improve the model, and the Grok Build harness gets better every day.
Nice work by all those involved!
Completely trained from scratch new models will be released by @SpaceX every month this year.
In the new video we merged 3 videos into one, a completely different format, and focused on the stories of the people competing. Feels very fresh, excited to keep experimenting 🙏🏻
Since 2008, $15B taxpayer dollars have been invested in California High Speed Rail. There is no working track and zero passengers so far.
In the same time frame $12B in private funds were invested in SpaceX pre IPO.
SpaceX has 10,000 Starlink satellites serving 10 million customers and has taken 74 passengers into orbit.
Politicians tell you they’ll do better with Elon’s money than he does. California shows the truth.
Starlink is connecting 100,000 students and 1,500 teachers in Malawi across 30 rural schools to online learning resources. For many, this is the first time students, teachers and families have access to reliable internet 🛰️❤️
Starlink V3 satellites have >10X bandwidth of V2 and there’ll be >10X launched, which means >100X more bandwidth.
Also, altitude will be 350km vs 550km, so min latency can be cut in half.
Light travels 300km/ms in space, so physics round trip min latency drops to <5ms.
SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible.
The potential speed improvement vs JAX for large training runs is over an order of magnitude.
Grok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come.
Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release.
This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.
If you're interested in how Artificial Intelligence works, and mechanical details of how Anthropic and others control it to improve safety, check out my new AI YouTube channel! I just made a video about sparse autoencoders, the main mechanism used to steer AI. https://t.co/EsukXkM06D