I ran the same Qwen3.6-35B-A3B at 256K context on an ASUS GX10 and a 64GB M2 Max. The main gap was prefill. The Mac needed 23–28 minutes before the first output token. Full results: 1. Prefill At 128K: GX10 TTFT 45.7s vs Mac 552.8s. Decode: 40.6 vs 28.7 tok/s. At 256K: GX10 127.5s vs Mac 1,541s median TTFT. Decode differs 1.4x. Prefill differs 12x. 2. MLX 4-bit beat 3-bit, 6-bit AND 8-bit 4-bit: 76 tok/s, 3/3 coding, 12/12 retrieval. 3-bit: failed executable coding in every single seed. 8-bit: half the speed and worse retrieval (mangled citations, 9/12). 3. MTP is not a free speed switch Same model, same Q4_K_M, same coding task, 3 seeds: GX10 CUDA: +26.9% decode. M2 Max Metal: −19.4%. Correctness held for both (12/12 executable) but speed didn't. Code: Raw dataset:

i added native support for @NVIDIAAI's Nemotron Puzzle 75B to mlx-lm. it now runs natively on an M2 Max 64GB: ⚡️22 tok/s 💾45.5 GB peak memory usage 📚4-bit experts + 6-bit dense + BF16 head i also fixed an annoying numerical bug in mlx-lm. outputs were subtly wrong, cosine similarity was 0.8832 vs NVIDIA's reference (identical inputs). the culprit was one dtype cast in the Mamba layers happening in a different spot than NVIDIA's. once moved, the cosine similarity improved to a satisfying level (0.999...). related PR: weights:

📊Today’s #BIT# Daily Chart - May 6, 2026 ⬇️ Volatility Is Compressing — Fuel for Bitcoin’s Next Leg Higher? #BIT# #Bitcoin# #BTC# #CryptoMarkets# #ImpliedVolatility# #MarketMomentum# #DigitalAssets#

🚨GPT-6 Astra Pro & Fable 5.1 are now above the human baseline on SimpleBench For a little bit of context this benchmark was literally built around the everyday reasoning tasks humans were still better at This gap has basically disappeared now, we aren’t far from AI outperforming even the smartest humans at (human) reasoning tasks

GPT-6 Astra >> Fable 5.1 , and it's not even an argument Artificial Analysis have updated their scoring methodology bumping Astra up a bit I still can’t see a world where this thing is below Fable 5.1, this is way off imo

This demo got a bit of traction last year so I was curious to see how much models have improved since then This is GPT-5 vs GPT-6 Astra making a pixel art solar system animation, the Astra one looks beautiful and the animation is really consistent too

Sol 5.6 Ultra gives surprisingly similar outputs every time, to the point where I've been investigating whether there's some illicit memory use thing going on... Don't think it's secret memory, just a bit RL fried + multiagent stuff collapses the randomness a bit (?)

Thought I'd share a bit more about the models I'm using together to build my first cybersecurity eval suite for @VulcanBench Starting with Opus 4.8 Medium for planning and the suite mvp build, then sending to GPT 5.6 Sol Medium for updates/polishing, then sending to Grok 4.6 High to perfect and really make sure they're good. Three step process, and yes, as you probably noticed, Opus 4.8, not 5, and I only use High with one model. Still enjoy Opus 4.8 for prototyping and conversing with, it's a fun model to ideate with on eval suites. But, it's far from perfect. GPT 5.6 Sol adds quite a bit of polish, and catches stuff that me and Opus miss. In the end though, I trust Grok the most, and it tends to always think deeply about some aspect of the eval suite Opus and Sol both missed. And Grok is totally honest with me about how good or bad my current eval suite is, and pushes me to do better all the time. note: please don't confuse this with the stack I use for coding. this is a different stack than what I code with. I share my coding stack, and how it changes in my substack (

OK, GPT-5.6 Luna is a bit of a beast. Given the 80% price drop today I decided to try it in Datasette Agent, and it's furiously quick and generates all the SQL, HTML and JavaScript (for Datasette Apps) I could possibly want

🏆 Round 6 is heating up. The final Marathon round before #STEPNathon# is bringing out the most consistent squads. Current top teams: 🥇 Achiever 🥈 ARPHA 🥉 StepnRU Every bit of Energy matters now, and some teams are making a serious final push before the finale. Is your squad holding strong or making a comeback? Drop your team name and current rank 👇