The world's largest GPU fleet, Cursor says, now sits behind the coding agent a lot of teams already pay for. SpaceX acquired Cursor so later models can be stronger and cheaper to run. Grok 4.6 was the early look.
Token Monitor puts AI coding limits on your desk.
The 4-inch screen tracks Claude Code, Codex, and Antigravity usage. Its maker says it reads token counts, not prompts or code.
750 output tokens a second on GPT-5.6 Sol. OpenAI says Ultrafast is up to 14× the same model's Standard tier. Until now, it says, real-time speed meant a smaller or more specialized model. Limited preview, on Cerebras.
4.6 to 28.3 on Terminal Bench 3.0, Z.ai says, from post-training only — same base as GLM-5.2. ExploitBench more than doubled, 24.4 to 54.4. Weights in two weeks.
