kefallrain @kefallrain

github.com/lepetitprince7… Joined October 2020

Tweets

46
Followers

7
Following

334
Likes

120

kefallrain @kefallrain

2 days ago

所以说，是astra赢了！

0 0 0 6 0

View Details

kefallrain @kefallrain

2 days ago

@Da7_Tech totally agreed

0 0 0 83 0

View Details

kefallrain @kefallrain

2 days ago

@Da7_Tech stardust-iota.vercel.app

0 0 0 4 0

View Details

🥹Just tell Opus to do its best — max effort for 2h 32m. Not quite suited for phone stardust-iota.vercel.app - Opus 5.5: 670 in / 4.7k out / 174.8M cache read / 829.8k cache write - Cost: $64.88 | API 152m | Wall 148m

0 0 0 23 0

View Details

kefallrain @kefallrain

3 days ago

@Da7_Tech Can't believe when Fable 5.5 drops. And I am waiting for K3.1, hope just don’t do the grok thing 🥲

0 0 0 396 1

View Details

kefallrain @kefallrain

4 days ago

@dabit3 Industrial porous carbon equipment interface and fine-tuning with agents

0 0 0 295 0

View Details

kefallrain @kefallrain

5 days ago

@RobertJBye It would be better to enable turning off ready for review and projectwithout deactivating it, since sometimes I just want to see the usage on the homepage instead of some long-lasting PRs and active project.

0 0 0 681 0

View Details

kefallrain @kefallrain

5 days ago

@Teknium Wait what

0 0 1 821 0

View Details

kefallrain @kefallrain

5 days ago

Both xhigh, same environment and tools, and turn 2 the cost was already $5.7🥹

0 0 0 7 0

View Details

kefallrain @kefallrain

5 days ago

Grok 4.7 token efficiency is not as good as 4.6, so Grok Heavy is no longer a good daily driver (it used not to be either). Even if I set auto-compact at 36% to avoid context beyond 200K, which works with 4.6, with 4.7 one prompt costs $5 just to write the plan.( and I also asked 4.7 to ensure the pricing for context window ≧200k haven’t changed yet?)

1 0 0 30 0

View Details

kefallrain @kefallrain

6 days ago

@Da7_Tech Fable can find out exactly what I want step by step, or help me out of my plenty of chaotic thoughts.

0 0 0 627 1

View Details

kefallrain @kefallrain

7 days ago

@amorriscode Far better than single long session to handle, very good improvement! If it can run locally, some local-specific tool calls can be handled easily, and a session timeline would be better than categories to manage super-long tasks from my perspective.

0 0 2 236 0

View Details

kefallrain @kefallrain

7 days ago

@orca_build Could we get a system change skin, and a transparent skin for terminal in desktop

0 0 0 122 0

View Details

kefallrain @kefallrain

a week ago

@Da7_Tech When they first rolled out GLM 5.2 and ran aggressive token-plan sales no one could buy manually in CN, a month later they “upgraded” the token plan with much higher cost, lower quota, and unbearably slow speed. I believe not many CN users would like Zhipu after GLM 5.2

0 0 5 607 0

View Details

kefallrain @kefallrain

a week ago

Oh no, I hit my Devin free-model usage limit on my Max plan 😰 even though I’m not a heavy user. I missed SWE2’s launch day. Nowadays SWE2’s speed is quite close to K3😭