Um achado interessante: Eu coloquei o OpenBot pra rodar todos os meus agentes com o Grok CLI usando o Grok 4.7 e 4.6
Os nรบmeros de um dia de trabalho:
Grok Bot: 196M de tokens
OpenBot: 119M de tokens
O Grok 4.7 entregou super bem o que eu precisava, com exceรงรฃo de uma รบnica aรงรฃo que envolvia a configuraรงรฃo de um worker do Cloudflare onde ele insistia que o problema estava no token e o Opus 5.5 rapidamente viu que era um erro do script.
No geral o que eu observei รฉ que TALVEZ o Grok Bot gaste token demais para fazer coisas medianas.
๐ BENCHMARKED: @thewebAI's TwIL-8B, an 8.8B open-weight reasoning model, just went 69/69 on my tests running on my 64GB M5 Pro Mac mini.
The 4-bit MLX build runs at 53.8 tok/s from a 4.9 GB file.
Full bf16 with the whole 128K context fits too, peaking at 40.5 GB on a 64GB machine.
Same scores on both builds, thinking on:
โ Math 20/20
โ Code 10/10
โ Instruction following 10/10
โ Harder math 8/8
โ Harder code 5/5
โ Tool calls 5/5
โ Hidden fact at 32K and 64K: 8/8
โ Hidden fact at full 128K: 3/3 (bf16)
4-bit speed by context:
- 512: 53.8 tok/s
- 32K: 26.6 tok/s
- 128K: 9.2 tok/s at 27.9 GB peak
bf16 speed by context:
- 512: 16.7 tok/s
- 32K: 12.6 tok/s
- 128K: 6.6 tok/s
@Davidstout called it "small enough to run on your own hardware."
I checked and it runs at full precision with the whole 128K context on a desk I own.
Images, upgradedโintroducing ChatGPT Images 2.6.
Now with GIFs and reference images.
Plus, two new models are coming to the API with 50% faster image generation.