Amit Kumar Jha

That's the front page of @Github Marketplace! 🎉🥳 cc @CirunHQ 🤌

You could mortgage a flat for that price in Manchester.

I do not understand, why would you have that as default.

@badlogicgames @threepointone

i'm sorry, @threepointone made me do it...

2-bit Qwen3.6-27B GGUF made 26 tool calls, triaged 15 GitHub issues and fixed, tested + reproed our repo’s 3 latest issues. 🔥 Try this locally in Unsloth Studio with just 12GB RAM. Studio also has a new look! GitHub: github.com/unslothai/unsl…

@badlogicgames @antirez @Kimi_Moonshot @mitsuhiko I have a little extension which adds support for cloudflare provider in pi, to be able to use any model from workers-ai including Kimi-K2.6

@antirez @Kimi_Moonshot export KIMI_API_KEY=key pi cltr + l kimi-for-cod + ENTER i'll make this more obvious in the next release (@mitsuhiko will)

been working with @Kimi_Moonshot K2.6 for the past 2 hours "accidentially" and ... it's really good!

Did the math on a @CirunHQ customer, ~6.5M CI-minutes last month on 16–64 core self-hosted GitHub Actions runners. At typical per-minute pricing (matched core count) by most providers, they'd have been billed $400K+. Their actual cloud + cirun bill was less than 1/10th of that. That's over $4M saved per year.

@Kimi_Moonshot M5 Ultras are going to be more expensive than gold, if they make it to the market.

Meet Kimi K2.6: Advancing Open-Source Coding 🔹Open-source SOTA on HLE w/ tools (54.0), SWE-Bench Pro (58.6), SWE-bench Multilingual (76.7), BrowseComp (83.2), Toolathlon (50.0), Charxiv w/ python(86.7), Math Vision w/ python (93.2) What's new: 🔹Long-horizon coding - 4,000+ tool calls, over 12 hours of continuous execution, with generalization across languages (Rust, Go, Python) and tasks (frontend, devops, perf optimization). 🔹Motion-rich frontend - Videos in hero sections, WebGL shaders, GSAP + Framer Motion, Three.js 3D. 🔹Agent Swarms, elevated - 300 parallel sub-agents × 4,000 steps per run (up from K2.5's 100 / 1,500). One prompt, 100+ files. 🔹Proactive Agents - K2.6 model powers OpenClaw, Hermes Agent, etc for 24/7 autonomous ops. 🔹Claw Groups (research preview) - bring your own agents, command your friends', bots & humans in the loop. - K2.6 is now live on kimi.com in chat mode and agent mode. For production-grade coding, pair K2.6 with Kimi Code: kimi.com/code - 🔗 API: platform.moonshot.ai 🔗 Tech blog: kimi.com/blog/kimi-k2-6 🔗 Weights & code: huggingface.co/moonshotai/Kim…

You can now visualize Pi traces that you upload on @huggingface! Let's make sharing agent traces 10x more common to make agent AI more open and collaborative! Also, because it's fun to analyze @badlogicgames's traces 😂😂😂

Now I just need half a dozen H100s

@mitsuhiko @fatihxkarahan It would be great to know what model I’m running when swapping local models on the same endpoint.

@fatihxkarahan Curious about the use case you have in mind.

If something really bothers you about pi, give us feedback please :)

@0xSero The delivery was pretty fast!

God damn mint fresh

Tried Muse Spark (from @meta's msl) just now for image editing, asked to convert the 3090 on the left to 5090, it came up with the image on the right.

Been trying Gemma 4 26B-A4B (MoE) for coding and its been interesting...

@badlogicgames Nice! I did something similar recently: github.com/aktech/pi-chan…

vibed a pi extension that lets you connect to a pi session via telegram. it's basically a mini claw. pi install github.com/badlogic/pi-te… it's kinda hilarious. 1k LOC, does all i need, including file attachments. trivial to setup.

Turns out $25 for a PR review isn't such great value when the reviewer leaks its own source code.

Ref: arxiv.org/pdf/2511.00739 newsroom.arm.com/news/arm-agi-c…

Everyone's been GPU-brained for 3 years. But agentic AI flips this. CPUs now account for 50-90% of total latency doing orchestration, tool calls, API handling, memory lookups. If you don't have enough CPUs, your GPUs just sit there waiting to be fed. Intel literally can't keep up with demand. AMD's calling it a surge. Meanwhile Arm just made its own chip for the first time ever. 136 cores, 300W, and Meta's already buying. The boring chip is suddenly the one everyone needs.

Step 1: share random shit with confidence Step 2: Then ask grok to verify

Europe and China are both investing heavily in RISC-V chips. Thanks to @openKylin the @condaforge project is making progress as well. Today I ran Pixi on RISC-V for the first time. If you're investing in RISC-V infrastructure and want the package tooling to match — let's talk.

Moved a few of the prod workers from AWS to Hetzner to save costs. Forgot the DB is still in AWS. Now the egress bill is higher than the VMs ever were.