a weird inversion with LLMs is the models improve faster than the tinkerers
when i see people with custom workflows and setups they're all addressing problems that don't exist anymore
the person naively using vanilla codex is more likely to be experiencing state of the art
HALO terminal retweeted
Overview of our recently launched AA-WER Streaming benchmark, measuring streaming Speech to Text models on accuracy and latency for voice agent use cases
Streaming Speech to Text (STT) powers real-time transcription in voice agents and live captioning, where models must balance accuracy against speed. Fast transcripts keep responses feeling natural and free up the response-time budget for reasoning and tool calls. Accuracy matters too, since errors can compound downstream.
Streaming STT models transcribe audio as it is fed in, sharing outputs continuously, unlike offline (batch) models that process the entire file at once and are typically slower.
Models from Cartesia, ElevenLabs, and Deepgram sit on the accuracy-latency Pareto frontier. Cartesia Ink-2 leads on final transcript accuracy at 3.59% WER (210ms), closely followed by ElevenLabs Scribe v2 Realtime at 3.64% WER (140ms). Deepgram Flux is fastest at ~20ms on final transcript latency (7.36% WER).
In this video, Kiriill Butler, Member of Technical Staff at Artificial Analysis, walks through the benchmark and key results.
HALO terminal retweeted
Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities
- Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas
- MiniMax Sparse Attention scales context to 1M
- Natively Multimodal from Step Zero
API: platform.minimax.io
Token Plan: platform.minimax.io/subscrib…
🚀New! MiniMax Code: code.minimax.io
Weights & Tech Report in ~10 Days
HALO terminal retweeted
Introducing MiMo-V2-Pro & Omni & TTS
mimo.xiaomi.com
HALO terminal retweeted
Introducing MiniMax-M2.7, our first model which deeply participated in its own evolution, with an 88% win-rate vs M2.5
- Production-Ready SWE: With SOTA performance in SWE-Pro (56.22%) and Terminal Bench 2 (57.0%), M2.7 reduced intervention-to-recovery time for online incidents to 3-min on certain occasions.
- Advanced Agentic Abilities: Trained for Agent Teams and tool search tool, with 97% skill adherence across 40+ complex skills. M2.7 is on par with Sonnet 4.6 in OpenClaw.
- Professional Workspace: SOTA in professional knowledge, supports multi-turn, high-fidelity Office file editing.
MiniMax Agent: agent.minimax.io
API: platform.minimax.io
Token Plan: platform.minimax.io/subscrib…
HALO terminal retweeted
Last Christmas at brunch, my friend Peter showed me the latest version of his little bot he built. A crazy side project. Just for fun.
This Monday he announced he's joining OpenAI. Sam Altman called him a genius.
In between? OpenClaw went from a hobby project to the most talked-about AI agent in the world. Three months. No team. No funding. Just one retired guy from Vienna who rediscovered the joy of building things. ❤️
Last week I flew down to Vienna to sit down with him right in the middle of this madness to talk about everything: how he thinks about agents, why he let OpenClaw write its own soul file, why your mom will understand AI agents before most engineers do, and why open source as we know it might be completely broken.
We are super proud of you, Peter. 🙌
Chapters:
00:00 Intro
01:02 The mom test moment & Managing 3,000 PRs 04:47 The Soul Markdown file & why OpenClaw feels different
06:58 The agent's personality, humor & OpenClaw's wild week
10:31 Think like the model
12:37 Privacy & security
15:28 Open source burnout & the future of engineering teams
17:32 The end of apps & closing thoughts
Day 6: OpenClaw almost bankrupted me overnight 🦞
I’ll admit it: I got too excited and a little cocky.
My AI agents nearly make me went bankrupt. 💸
The "heartbeat" trap is real:
- 9 agents pinging all night obliterated my wallet.
- GLM-4 glitched, fallbacks triggered, and costs exploded while I slept.
I’m 25. I can’t afford idle brains or token bloat.
So I burned it down and reset to zero.
I stripped the heartbeats and simplified to a 2-agent elite squad.
The result? The system is faster, the logic is cleaner, and the overhead is gone.
I’m not just building a startup.
I’m building a lean, mean execution machine.
Sometimes you have to break the system to learn how to master it.
Back to building.
HALO terminal retweeted
Spannende Erkenntnis aus unserem Blickwinkel. Die lautesten KI-Kritiker kommen selten aus Tech. Sondern aus dem mittleren Management. Die höchste Dichte an Widerstand bei gleichzeitig geringster Tool-Nutzung.
HALO terminal retweeted
I’m still processing this shock 😮
Opus 4.6 just migrated my entire website, 456 pages of content across multiple categories, from WordPress to Jekyll in one shot.
this felt like watching an industry category boundary collapse in realtime 😵.
That was the problem I had as well. But I went for building my own hardware solution. Lights up blue if idle/done, yellow while working, and red when permissions are requested:
Replying to @tonysheng
My current workflow has 6+ tabs of Claude Code running at all times.
M pain: I'd leave a tab unattended, come back 20 minutes later, and realize Claude had been waiting for permission the whole time.
I also would have to navigate from tab to tab checking all. Felt broken.























