Scerw it. I'm goimg to say it. I like minimax 3.1 flash more than deepseek v4.1 flash. Its just too good with terminal stuff.
It’s like a coordinated attack against opencode. I don’t get it. It started with @NoemiTitarenco and now I see at least 2 or 3 of these every time I open twitter.
Why is a similar comment to this showing up on my TL every time I open X?!
opencode で提供されている DeepSeek-V4.1-Flash-1017 は量子化により劣化しているのでは？という言説がTLで流れてきたけど、彼らは公式でリリースされているモデルは最初からほぼ MXFP4（EngramはFP8） で構成されているという事実を知らないのかな
本当にDeepSeek 公式APIと挙動が異なるなら、DeepSeek公式だけが量子化していないバージョンを持っている可能性もあるけど、API料金が同じ（オフピークはむしろ公式のほうが安い）ことを考えるとおそらく同じなのではないかと推定できる
ちなみにAnomalyの共同創業者でありopencodeを開発しているDax氏は、 opencode では公式の提供しているのと同じモデルを展開していると述べている
Happy to see @nothingto12 getting some traction from opencode and command code. He’s doing god’s work.
Paka Joe retweeted
QoderCN 的deepseek 4.1 flash 也被实锤会路由到qwen3.8系列
速度明显过慢，而且ai分析指纹偏向qwen3.8家族
白嫖就好，一言不和就排队，果然掺水了，别付费
Paka Joe retweeted
$52M seed round. 15M+ users. But this is what we're most proud of yet.
Introducing Drama 3: the most emotive and controllable TTS model in the world
And yes, it's open-weight.
You can now control how even a single word should sound, change its tone, and shift your voice mid-sentence.
Uncle Roger was impressed too.
Try it here:
s.fish.audio/rqh0l8
Comment “DRAMA” within the next 48 hours for 1 month free.
If you're an enterprise your first month is on us:
s.fish.audio/qz9r0t
Is it better to use "herdr agent wait" or a loop when having one session drive another? I like agent wait simply because I can tell that it's still working.
Solid evals from @favurdev. I dug into the scoring methodology today (okay, my agent did), and this is genuinely useful for picking models by role. Good site to keep in your back pocket.
Favur Evals - weekly recap.
We rescored every run this week. A quarter of each score is now whether the final build was verified complete, so your model's rank may have moved.
Same task for every model: build one window with a circle orbiting a circle.
@StepFun_ai 's Step 5 Preview finished more of the Python build than any model we have run on it.
@OpenAI 's GPT-6.1 Sol wrote the highest-scoring TypeScript test suite, with every test passing and every line covered.
@Zai_org 's GLM 5.3 Prime built it in Java, C#, Rust, Go and JavaScript, and in each language it ranks first or second for fewest tokens and turns.
new recap every Friday ↓
evals.favur.dev/go/x/evals-r…
Paka Joe retweeted
Step 5 Preview is now live on OpenRouter
I gave 4 models the same 3 tasks: build an Eiffel Tower, a rocket launch, and a building collapse all in Three.js
Step 5 Preview came out cheapest and fastest across the board and got good results not expecting this from an open-weight model at all. DeepSeek V4 close behind. GLM 5.3 and Qwen 3.8 took longer and cost more, but still landed solid results.
Video says more than I can.
Happy to see some real step 5 results coming in. I've been searching for weeks since they dropped it on their own inference.
Many non-OpenAI and Anthropic models dropping lately, but @XiaomiMiMo v2.6-pro has been the most reliable workhorse for me.
