hard to keep deepseek down they are back on top with 6.6T tokens for the day

Sep 12, 2026 · 4:05 AM UTC

95
45
2,048
178,495
RepliesRelevant
Replying to @opencode
@opencode Because 4.1 Flash is genuinely a really good model.

I tested it on the LLM coding benchmark and it was the best model to generate frontend code, as judged by GPT Sol.
1
1
49
4,835
Replying to @opencode
@opencode deepseek is cooking hard
0
0
2
868
Replying to @opencode
CoreWeave Engineering might have discovered a ✨problem✨
0
0
0
3,477
Replying to @opencode
@opencode Can we have the $60 usage for a bit longer?
1
0
75
7,000
Replying to @opencode
@opencode deep seeks the only LLM ive been hard for
0
0
14
4,014
Replying to @opencode
@opencode Enjoy while you can. Amodei and Scam Altam are doing all they can to ban deepseek.
1
0
13
2,618
Replying to @opencode
@opencode Wow! Even though Meta is fully free
1
0
10
2,940
Replying to @opencode
@opencode Did opencode 2 just come out of beta?
0
0
8
2,652
Replying to @opencode
0
0
5
1,733
Replying to @opencode
@opencode Chinese ones re really cheap and great also, idk why Americans does not have cheap ones
1
0
4
2,409
Replying to @opencode
@opencode Why even try to keep them down? They have one of the best offerings that strikes a sweet spot
0
0
2
1,222
Replying to @opencode
@opencode Seems to be a fan favorite. Very fast and 'just' good enough.
0
0
2
2,465
Replying to @opencode
@opencode Who tf is using a meta ai? I call bullshit
0
0
1
688
Replying to @opencode
@opencode May I ask how do you ensure that they are ZDR when they are hosted by the DeepSeek in Hangzhou, China? I thought they are legally required to retain data
0
0
0
19
Replying to @opencode
@opencode Somehow, DeepSeek fails for me a lot. Somehow, I cross its restrictions, and I have no idea how
0
0
0
963
Replying to @opencode
@opencode They will be down as soon as the 4x usage expires. IMO, their usage must be increased to at least 30 dollars.
0
0
0
2,151
Replying to @opencode
@opencode Does this data include free users, or is it only for Go users?
0
0
0
240
Replying to @opencode
0
0
1
734
Replying to @opencode
@opencode Same day, Meta is at 6.5T. 39.7 against 40.1.

That is a 0.4 point lead on a 16T day.

Those two are almost 80 percent of the whole thing. Third is Xiaomi at 1.7T.
0
0
0
832
Replying to @opencode
@opencode everyone loves cheapseek
0
0
0
225
Replying to @opencode
@opencode Deepseek is unstoppable
0
0
0
332
Replying to @opencode
@opencode 6.6T tokens is a weird way to stay down
0
0
0
840
Replying to @opencode
@opencode can we have 60$ ds v4.1f forever?
0
0
17
1,523
Replying to @opencode
@opencode DeepSeek carrying 40% of that chart solo, everyone else just visiting
0
0
2
285
Replying to @opencode
@opencode 4.1 Flash is a really good model, but I don't get why Zhipu is only at 3.9%, imho their models are often times even better with a similar price tag.
0
0
2
1,122
Replying to @opencode
@opencode 可以保持长期的6x吗?我认为这是OpenCode Go的核心竞争力。
0
0
1
630
Replying to @opencode
@opencode Last week, there are 22T coding token in openrouter. Opencode make 16T in a Day 💪. That is why openrouter no longer track opencode usage.
0
0
1
231
Replying to @opencode
@opencode Impressive usage. Open models win when latency, cost, and quality align.
1
0
1
299
Replying to @opencode
0
0
1
84
Replying to @opencode
@opencode Can @grok separate a real usage rebound from a single benchmark spike when token counts jump this fast?
1
0
1
1,255
Replying to @opencode
@opencode Can't beat deepseek flash in terms of value for money

Exploring the whole @opencode codebase + drafting a software diagram with code reference costed 0.2$ in tokens using buildsee.dev
0
0
1
116
Replying to @opencode
@opencode 你们的 v4.1flash 慢的像吃屎了一样,你们这群傻逼到底在干嘛????不想干就别你妈干了
0
0
0
22
Replying to @opencode
@opencode The universal law of AI: "everything will be whale in due time."
0
0
0
32
Replying to @opencode
@opencode C 端要的是对话手感,不是写代码吞吐。接口名还叫 Pro、底下静默换 Flash,本来就会炸。改主意比硬切体面,路由表写死 Pro 的可以先松口气。
0
0
0
357
Replying to @opencode
@opencode too bad it sucks at coding
0
0
0
12