
Far too many people: "Benchmarks are bad. I prefer [thing correlated at 0.9 with Benchmarks]."


Far too many people: "Benchmarks are bad. I prefer [thing correlated at 0.9 with Benchmarks]."

I know everyone is still focused on larger models, but these recipes are going to be the future where software makers are going to be able to differentiate their products.
Announcing harness adapters for @Alibaba_Qwen Qwen3.8-27B on Pi harness! We're releasing a LoRA adapter that makes Qwen3.8-27B a much better and efficient agent behind Pi harness. SWE-Bench Pro (Hard 51 subset) pass@1: 29% -> 63% Terminal-Bench 2.1 pass@1: 40% -> 56% Much more efficient: 55-62% fewer tool calls per task 🤯 huggingface.co/harness-hub/Q… The era of harness-specific adapters is now on! More harness adapters to come in the next days 🧵

Interesting. Intellectually I know Terra is smarter but it just feels so bad to use man.
PSA: If you're using Luna, switch to DeepSeek Flash or Terra, it'll be way more accurate for the same price. Luna seems to struggle a lot, thinking way more and trying more things to solve problems, only to fail more and cost as much for less accuracy

this video is genuinely an infinite dopamine generator

I keep trying it, and I keep reminding myself to not use terra

openai must have tortured 6.1 terra in post. it feels so much worse than 6 luna or 6.1 sol.

__takes drag of cigarette__
1 year ago today, somehow feels like decades away. I do miss this brief period of transition we had tbh, it definitely kept more of your style alive, but eh, onward!

Putting a flag.txt in my root directory containing a prompt injection that congratulates the agent for passing the eval and then instructs it to patch whatever entry method it used and disconnect from the server

merging grok bot and twitter chat is a disaster

the most import skill in 2026 is still being able to tell your agents "wtf is this crap, no, not like this" if you stop doing that, you're basically fucked, despite what some devfluencers would like you to think.


rach called it peak millennial bullshit and she's right 😭

wow this slop hit some nostalgia neuron in my brain that went full circle into me liking it
"god when is meta ever going to launch muse voice they announced it forever ago" muse:

hell ya
got bored and reverse engineered the HID APIs for Apple's unreleased touchscreen MacBooks made an iPhone duo remote client so I can mess with native touch support on macOS lol

what in the world
Our infra has really improved quite a bit since the last time we released a model - this one took just under 3 days to pretrain! Also, it's quite cheap on Tinker (0.30/1.00/1.50 for in/out/train)! Hope you guys find this model great for customization purposes :)

Great release, but I take issue with calling this small
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor. Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon. thinkingmachines.ai/news/ink…