Andrey Baksalyar
253 posts
Andrey Baksalyar
@baksalyar
Telegram — @baksalyar, Instagram — @baksalyar
Ayrshire, UK Katılım Haziran 2011
722 Takip Edilen28 Takipçiler
@CommandCodeAI New GLM is coming? I noticed they're updated their webUI today. ;)
English
The next Command Code deal drops. Worth the wait.
- Monday June 15, 2026
- 10 AM PT
You'll want to be online for this.
English
stripe still has no dark mode on the dashboard in 2026
dev first company btw
opened it at 12am and got absolutely flashbanged
retinas are cooked lmao fix your shit @stripe
English
@MiniMax_AI M3 is awesome! Join the party (not the one you thought of!) and snag a 10% discount with my referral link:
platform.minimax.io/subscribe/toke…
English
M3 would never 🙂↔️
As a matter of fact, the weights are now open, too.
huggingface.co/MiniMaxAI/Mini…
Anthropic@AnthropicAI
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…
English
@ArtificialAnlys How long do you think it'll take to run all the models through the Deep SWE Bench again? I'm really curious to see how things shake out in the second tier once the Chinese open-source models enter the picture.
English
Today is the first time our Intelligence Frontier chart has moved backward.
English
@CommandCodeAI I asked it to find bugs in the codebase — no results in the end (exploration got interrupted) and -$5 drained my account credits 😅 I know it's not real money in cmd, but still. It doesn't make any sense if it's no better than GPT 5.5.
English
Kimi K2.7 Code is now in available in Command Code.
10x free credits in Go.
Our new #1 open mode in internal benchmarks.
cmd update to v0.37.0
select via /model
• 256K context 🍃
• 30% lower reasoning tokens than K2.6 ✅
• Open weights 1T-parameter MoE - 32B active ⚡
English
@MrAhmadAwais I asked it to find bugs in the codebase — no results in the end (exploration got interrupted) and -$5 drained my account credits 😅 I know it's not real money in cmd, but still. It doesn't make any sense if it's no better than GPT 5.5.
English
ok kimi k2.7 code seems like a real deal.
it's already dethroned k2.6 in every test i've thrown at it.
still benchmarking, but my early money is on this being the new top open model.
Command Code@CommandCodeAI
Kimi K2.7 Code is now in available in Command Code. 10x free credits in Go. Our new #1 open mode in internal benchmarks. cmd update to v0.37.0 select via /model • 256K context 🍃 • 30% lower reasoning tokens than K2.6 ✅ • Open weights 1T-parameter MoE - 32B active ⚡
English
I kept spotting typos in my Claude Code prompts after I'd sent them. So I built tuipo — Grammarly for your terminal
It underlines typos as you type in any TUI — CC, Codex, Aider, vim, your shell — and never touches the app it wraps.
github.com/ARahim3/tuipo
@steipete @bcherny
English
English
@luckyPipewrench @theo I wonder, what prompt caused this message?
English
@dillon_mulroy @beginbot Moreover, he sometimes does not disdain even Perl!
English
@harisamjed @MrAhmadAwais The same experience. It's good for big codebase scans though — searching for bugs, security holes, and issues with extensive reports, which are later rechecked by smarter models — ideal tasks to leverage its speed.
English
@MrAhmadAwais Well for some reason nemotron never worked for me. Yesterday I fell for Ultra version hype and gave it a complex task. It burned $4 of credits and messed up several files. I interrupted it mid task and change model to dsv4flash which fixed everything.
English
@ArtificialAnlys Where is MiniMax M3? Why did you ignore it? Are you pretending you didn't notice the release of one of the top Chinese LLMs? 😄
English
Alibaba's Fun-Realtime-TTS takes the #1 spot on the Artificial Analysis Speech Arena Leaderboard, surpassing Google's Gemini 3.1 Flash TTS and Inworld's Realtime TTS-2 Research Preview
Competition at the top of the TTS Arena is tighter than ever, with just 24 Elo points separating the top five models. Fun-Realtime-TTS takes the top spot with the highest Elo score on the leaderboard.
@Ali_TongyiLab @AlibabaGroup's previous Fun-Realtime-TTS-Preview reached #7 on the leaderboard, making this Alibaba's first #1 model in the Artificial Analysis Speech Arena. Fun-Realtime-TTS is available via Alibaba Cloud with API access for developers.
Key takeaways:
➤ Quality: Fun-Realtime-TTS has an Elo score of 1,219 (+16/-16) based on 962 arena appearances, placing it ahead of Gemini 3.1 Flash TTS at 1,214, Inworld Realtime TTS-2 Research Preview at 1,209, and Cartesia Sonic 3.5 at 1,203
➤ Pricing: Fun-Realtime-TTS is priced at $27.59/1M characters, positioning it between Gemini 3.1 Flash TTS at $18.3/1M characters and Inworld Realtime TTS 1.5 Max at $35/1M characters, while remaining below Sonic 3.5 at $39/1M characters.
➤ Features: Fun-Realtime-TTS supports real-time speech generation with voice cloning, voice design, multilingual output, and support for regional accents and dialects.
See more details and listen to samples below 🧵
English
@OxxoTweets So what does that even mean? 38T of garbage tokens and synthetic pretraining slop just to inflate the numbers and pretend the model has some "unique level of intelligence"? I tried it, and it couldn't even write a simple Python script — it got stuck on f-string formatting, lol.
English
@NiP73 What does it change? This thing can’t even produce a simple Python script after multiple failed attempts and can’t use tool calls, lol. Are you all insane? Wake up—it’s junk.
English
This needs to be massively celebrated.
Ignores the doomers that say cost is out of control. This changes everything.
Liquid AI@liquidai
Today, we're releasing LFM2.5-8B-A1B, a device-optimized model designed to power real-life applications on phones, laptops, PCs, robots, and fast & lightweight server-side use-cases. > 8B MoE, 1.5B active > Expanded 128K context > LFM2.5 flagship hybrid MoE architecture > Trained on 38T tokens + large-scale RL > fast, reliable tool calling, punching above its weight, comparable to models with up to 4x its size > customizable on a single GPU for any specialized task > LFM2 open-weight license 🧵
English
@Codedigipt Yeah, sure, “great performance”. It's a trash, man, it can't code it can't use tools, it hallucinates AF.
English
@NeoAIForecast @liquidai @AMD Man, it can't even make a simple, working python script after multiple attempts. What “much capability” are you talking about? It's just crap.
English
Benchmarking the new LFM2.5-8B-A1B MoE by @liquidai on an @AMD RX 7800 XT using ROCm and llama.cpp.
A few interesting results:
>Q5_K_M delivered the fastest generation speed at ~200 tok/s
>Q8_0 pushed over 6,000 tok/s prompt processing
>Even a 16GB card can run these smaller MoE models comfortably
>Shows just how much capability is available without needing flagship hardware
Running more tests on the AMD stack and seeing how far ROCm has come
English