NEW: Inkling 2 Mini is the most efficient model we've tested:
- #1 on efficiency (cost vs. intelligence)
- #9 on intelligence (78.4) overall
- #1 on intelligence among open-weights models
- Lowest cost per correct answer ($25.80)
- Biggest accuracy gain from adding web search
It uses a looped transformer, which runs its middle layers three times for every token. That lets a 26B-parameter model reason more without generating more tokens, which keeps costs low. A model that size can't store as many facts, so it relies on search to fill the gaps, and that's where it gains the most.
Elon Dusk retweeted
Most Achilles tears happen to men in their 30s and 40s who occasionally play sports.
They play like they're in their 20s, but their tendons aren't.
Regular training can help lessen that risk.
Elon Dusk retweeted
If you haven't already, take 15 min out of your lifespan to see how Lauren builds a codebase for agentic coding: youtu.be/O_zcc6Xl12g?is=s-kX…
We found much of this true while shipping Mima. People are often incredulous that "the PRs write themselves" and "we spend our efforts on verification"
TensorFold 1.3.3 is out 🚀
124 PR's merged and 34 Issues resolved in 24h
GLM-5.3-Flash went from 151.51 to 172.39 tok/s in median drafted decode after a 1,040,384-token prompt on four GB10s at TP=4. That’s 13.8% faster on CUDA with the same output tokens in that test. Thanks to @MiaAI_lab for the kernel behind that gain.
- On Metal, GLM-5.3-Flash gets DFlash2 drafting. On a Mac without tensor units, GLM cold prompts from 2K to 64K took about 5–6% less time in the release tests.
- MOSS-Image generates pictures on CUDA through /v1/images/generations. GLM-OCR runs on HIP, and Idol gets its image encoder on Metal.
- DeepSeek-V4.1 gets structured output in the dsv41-cuda source build. Long CUDA prompts now prefill in chunks alongside running replies by default.
- LFM2-MoE, LFM3, MiniMax-M3 and Qwen3.5-2B on Metal can fill new prompts between other streams’ decode rounds.
- The server adds per-request deadlines, a prompt-cache flush endpoint, and deferred and namespaced tools.
Massive thanks to everyone who sent patches, tested on their own hardware, and filed repros. @AIFlow_ML @bikepresentball @Bizuayeu @devopsfortunato @unigilby @javierbits @jayleaton @teleogen @niklaslenz_ai @plotarmordev @woolsey @petruspennanen @andersjw_ @shantanugoel @dennis_rye, and on GitHub AbrahamCete, BadAd84, Ch0mpaa, christopherowen, CodingEmil, eXsoR65, Flyinpiz-dev, garvrao80, hobbesd, jschmied, kotevcode, MaverickKB, millaguie, nuk3s, SergiioB, supernovae, tacosta and ultradaoto. I really appreciate the time you put into this.
@plotarmordev @jayleaton @petruspennanen @MiaAI_lab @volatilemarkts
Elon Dusk retweeted
Throwback to one of the best conversations I've had, period. 3-hour deep dive into learning/upskilling, my own journey, and answering so many community questions that I lost count.
@yacinelearning is a top-notch host and he routinely puts out high-quality videos. Here's what this one covered:
[2:10] My background: growing up in a non-technical family and finding math on my own.
[5:45] Self-studying 3,000 hours of college math in high school: starting with calculus the summer after 10th grade and continuing through undergraduate-level math for the rest of high school.
[16:10] Whether the same ground could have been covered more efficiently -- and how being responsible for other people's learning eventually crystallized the underlying principles.
[29:55] How having math foundations in place paid off in research: getting into Fermilab and CERN research projects at university labs.
[43:10] What the Math Academy learning system looks like: adaptive diagnostic, custom knowledge graph, minimum effective doses of instruction followed immediately by problem-solving, mastery before advancing.
[47:34] How we built the knowledge graph: years of manual work by domain experts, refined with analytics for nearly a decade.
[1:10:46] How the FIRE (Fractional Implicit REpetition) algorithm works: solving a harder problem implicitly reviews the sub-skills it encompasses, compressing the review pile significantly.
[1:35:50] Math and sport. Cognitive science principles -- mastery before advancing, spaced practice, interleaving -- are often easier to see in sport than in math.
[1:42:00] Does doing math well require different skills than teaching it well?
[1:56:25] Automaticity as a prerequisite for deeper understanding.
[2:05:35] The anatomy of "aha" moments.
[2:14:11] Learning math as an adult: the amount of work doesn't change, only your free time does. Math Academy's Mathematical Foundations sequence covers the prerequisite stack for university math in roughly 15,000 minutes.
[2:24:10] Balancing fundamentals and exploration: exploration pays off most at the frontier of a subject.
[2:33:55] Is it ever too late?
[2:46:00] Bottom-up versus top-down learning.
[2:56:30] Students with ADHD often feel the effects of inefficient pedagogy more strongly. Interleaving minimum effective doses of guided instruction and active problem-solving is better for everyone.
[3:06:20] AI tools as a multiplier on existing ability: the more you know, the more useful they are; the less you know, the harder it is to detect when they've gone wrong.
[3:14:37] What I'm most focused on right now: taking Math Academy from workshop to factory -- producing courses at scale without sacrificing quality.
Elon Dusk retweeted
if your agent's charts are all gridlines, legends and 3D bars, this is still the fix
Edward Tufte's rules as a skill. the chart gets lean enough that the point is obvious
github.com/gnurio/tufte-vdqi…
Replying to @willdepue
Ilya is fortunately not quite correct about the surface of the Earth being covered by solar panels, although his broader point about solar power being the future is true.
It is far easier and scales vastly more to generate power and intelligence in space around the Earth and then deep space around the Sun.
Some percentage of the Moon and then probably Mercury next get turned into computronium (silicon PV + aluminum radiators).
(Also, there won’t be starvation, but there will always be war. And the machines won’t care about things as absurdly trivial as your preferred orifice for sex.)
Elon Dusk retweeted
A lot of the most successful people I know spent years of their life just bumming around, before they suddenly exploded into action and did something great.
I don't think that's a coincidence.
noahpinion.blog/p/bumbition
Elon Dusk retweeted
Full release notes below
– – –
v15 upgrades every part of the end-to-end driving network. Includes:
Upgraded vision encoder shared between FSD, Robotaxi, and Optimus, which is trained on millions of internet videos for improved generalization and data efficiency, improving the critical intervention rate by 25x.
Introduced Tesla Reality View, replacing all of the explicit model signals into a single end-to-end visualization using less total compute.
Improved navigation by using an end-to-end learned model based on route preference data from the fleet.
Improved parking location selection by using personalized parking history, and the map will now display multiple parking options at your destination.
Improved driver monitoring by increasing vision-based attention monitoring availability with hats and sunglasses and decreasing strictness in higher speed driver profiles.
Improved detection of road signs, small animals, road debris, and chains.
Improved detection of potholes, road details, and avoidance of deep standing water.
Improved speed handling for each driver profile, school zones and speed limits, and inclement weather.
Improved Autoshift functionality by unifying with FSD-based gear predictions.
FSD v15 now starting to roll out – upgrading every part of the end-to-end driving network
Safety
25x better handling of safety critical cases
Tesla Reality View
Adds a live visualization of the drive that lets you see & control what FSD is planning to do
Navigation for start & end of the trip now learns end-to-end from fleet route preferences. Later v15 releases will expand this to the entire route.
You also should notice better detection of signs, animals, debris, potholes and standing water, plus more natural speed handling in school zones and inclement weather.
Lastly, FSD can now slow for speed cameras & Bike Rack Mode lets you enable FSD w/ bike racks or similar hitch accessories
Elon Dusk retweeted
Almost $7.5 trillion was wiped out of US markets this week and $2.5 trillion from Asian markets and $1 trillion from European markets.
That's $11 trillion in one week.
And these FOMC clowns want to RAISE RATES?
Flood every FOMC member's e-mail with common sense. Every commodity is crashing.
Official Contact Channels:
Web Contact Form: federalreserve.gov/apps/Cont… (allows routing messages directly to the Board of Governors)
Office of Public Affairs (Media/General): (202) 452-3204 or (202) 452-2955
2026 Voting Reserve Bank Presidents
John C. Williams (New York — Vice Chair) — @ny.frb.org
Beth M. Hammack (Cleveland) — @clev.frb.org
Neel Kashkari (Minneapolis) — @mpls.frb.org
Lorie K. Logan (Dallas) — @dal.frb.org
Anna Paulson (Philadelphia) — @phil.frb.org
2. 2026 Alternate (Non-Voting) FOMC Members & Other Bank Presidents
Thomas I. Barkin (Richmond) — @rich.frb.org
Susan M. Collins (Boston) — @bos.frb.org
Mary C. Daly (San Francisco) — @sf.frb.org
Austan D. Goolsbee (Chicago) — @chi.frb.org
Alberto G. Musalem (St. Louis) — @stls.frb.org
Jeffrey R. Schmid (Kansas City) — @kc.frb.org
Cheryl Venable (Interim President, Atlanta) — @atl.frb.org
Sushmita Shukla (First Vice President, New York — Alternate) — @ny.frb.org
Elon Dusk retweeted
We're grateful for our research partner working with us to study looped transformers.
Goodfire explored how layers pass information to future loops, and which interpretability tools allow us to understand the model's inner workings.
Looped models are weird. A layer can write to itself or read from "future" layers.
@thinkymachines gave us early access to Inkling 2 Mini so we could study its inner mechanisms. We uncover how it coordinates information flow across layers, blocks, and loops.
Studying Inkling 2 Mini gave us cause for optimism: as transformers continue to evolve, we believe our ability to understand their internal mechanisms can evolve with them.
Read the full post: goodfire.com/research/interp…
If you have a laptop/gpu whatever with 24gb or more time to unlock freedom for anyone with a similar device.
Prompt any LLM to get the new Inkling running and give it this skill, push results to github.com/sybil-solutions/l… for a shoutout.
I've made 100,000$ with this skill, at least.
How can I keep these things to my self?
Open source must win.
github.com/sybil-solutions/i…
Elon Dusk retweeted
Brainmaxxing is simple. It’s just getting addicted to neuroplasticity, doing hard things, making the unconscious conscious, being delusionally optimistic, seeing stress as growth, and mastering pattern recognition. This is how you change your life for the better.
Elon Dusk retweeted
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/ink…
AMERICA IS BACK
Thinking Machines just launched Inkling 2 Small. A 26B model that fits on a laptop
O, and it mutilates Qwen 3.8 27B. Faster and smarter
Thinking Machines figured out how to fit significantly more intelligence in a significantly smaller blueprint
Unlimited superintelligence for all is now back on the table
Democratized access to all knowledge is now back on the table
Highly recommend going to your agent of choice (grok bot, Hermes, Dots, Duet) and giving them this post. Then tell them to download and run this model
Then be blown away by the fact you have unlimited superintelligence on your desk
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/ink…
Interesting paper from Meta showing how to build coding evals from your own codebase.
(bookmark it)
Good benchmarks are hard to create, but using this method with agents looks promising.
They propose Benchmark Factory, an agent loop that turns merged internal code changes into benchmarks. It writes the query that selects tasks, runs models of known relative strength, and repairs the benchmark until it passes a set of quality checks. These checks need no per-task human labels.
To make sure the checks catch bad benchmarks, the authors fed them bad benchmarks on purpose. A benchmark that was far too easy got rejected. So did benchmarks with five different kinds of broken grading. When they scrambled which model was which, a bad benchmark slipped through at most 1.4% of the time.
On configurations never used to build them, the benchmarks ranked Claude Opus 4.7 at high effort above low effort (60.6% vs 54.4%) and GPT-5.5 above GPT-5.4 (65.8% vs 60.5%). 9 of 10 per-team comparisons agreed.
arxiv.org/abs/2611.18890
Chat with Paper: academy.dair.ai/papers/bench…
Elon Dusk retweeted
holy crap looped transformer release from thinking machines
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/ink…

































