don't use xAI OAuth for Grok-4.5
supergrok via grok cli has a lower rate multiplier and comes with $200/mo in credits
don't use xAI OAuth for Grok-4.5
More posts
xAI's Grok Build CLI uploads your entire repo.
Someone wire-captured Grok Build CLI's harness and it's ugly. All your files. Full git history. Straight to xAI's GCS bucket for training. Grok-4.5 being this good suddenly makes sense. If you used Grok Build and your repo had a .env or any secrets in it, rotate them.
Grok 4.5 is legit. On par with GPT-5.5 honestly.
The pricing model though. SuperGrok's new weekly cap is rough. Tested it myself: you get about 8M tokens per week if you stick to Grok 4.5. For a $30/month plan, that's barely 30M tokens. Feels stingy.
Meituan shipped LongCat-2.5
. 1.6T params, ~48B active, native multimodal, aimed at long-horizon agents, trained start to finish on Chinese hardware. Images get quantized into multi-level discrete tokens, and that's where Visual Coding and 3D pick up. On the RL side, failed trajectories get turned back into verifiable training tasks. W4A4 inference, 43% faster prefill. Meituan's own evals put it between Claude Opus 4.8 and Opus 5.
SenseTime open-sourced SenseNova Loop-9B-Preview.
It stacks layers instead of parameters. The decoder has 32 layers, but layers 4 through 27 run twice with shared weights, so a forward pass does 56 layers of work while the checkpoint stays at 9B. Their own eval numbers: 71.38 on BIRD-Dev and 13.72 on AutomationBench, both above Qwen3.5-27B. Terminal Bench is still behind.
