Data centers dominate AI, but they're hitting physical limits. What if the future of AI isn't just bigger data centers, but local intelligence in our hands? The viability of local AI depends on intelligence efficiency. To measure this, we propose intelligence per watt (IPW): intelligence delivered (capabilities) per unit of power consumed (efficiency). Today’s Local LMs already handle 88.7% of single-turn chat and reasoning queries, with local IPW improving 5.3× in 2 years—driven by better models (3.2×) and better accelerators (1.7×). As local IPW improves, a meaningful fraction of workloads can shift from centralized infrastructure to local compute, with IPW serving as the critical metric for tracking this transition. (1/N)

super sick stuff on fast local models from @jackyk02, @JonSaadFalcon, @Azaliamirh, @HazyResearch + team

System 1 models have generated significant excitement recently, particularly following the introduction of #Jev. We introduce Contrastive Language Model (CLM): an ultra-fast System 1 model that connects states and actions through contrastive learning. CLM-8B delivers up to 9× faster inference than Jev ⚡ while achieving comparable performance across computer use, gaming, and tool calling. With lightweight fine-tuning, it achieves 81.6% on DeepSWE and 87.6% on Terminal-Bench 2.1. We also introduce efficient training and serving infrastructure for CLMs and establish scaling laws across compute, model size, and dataset size. 📄 Blog: contrastive-lm.notion.site 💻 Code: github.com/Contrastive-LM/CL… 🗣️ Discord: discord.gg/5dAQEDJBs 🤗 Data & Models: huggingface.co/Contrastive-L… Work led by @jackyk02, in collaboration with @hangoo_kang @TarunSures41845 @JonSaadFalcon @Azaliamirh @HazyResearch More details in @jackyk02's post: nitter.net/jackyk02/status/210290… @StanfordASL @StanfordAILab @StanfordEng

Introducing Contrastive Language Models (CLMs), a System 1 model that connects actions and states! Agentic coding: with lightweight finetuning, CLM-8B sets a new SOTA on DeepSWE (81.6%) and Terminal Bench 2.1 (87.6%). CLM-8B vs. Jev: comparable zero-shot performance on computer-use, gaming, and tool-calling tasks, while being up to 9× faster. We are releasing the CLM-8B checkpoint, its data, and infra today!

We’re pleased to announce our work Contrastive Language Model (CLM), an ultra-fast System One Model trained with contrastive learning for Internet-scale data. CLM matches or outperforms Jev at substantially lower latency (up to 9x). Exciting work with @jackyk02 @TarunSures41845 @JonSaadFalcon @drmapavone @Research_Hazy @Azaliamirh 🌟!

I’m pleased to announce our work Contrastive Language Model (CLM), an ultra-fast System One Model trained with contrastive learning for on Internet-scale data. CLM matches or outperforms Jev at substantially lower latency (up to 9x). Exciting work with @jackyk02 @hangoo_kang @JonSaadFalcon @drmapavone @Research_Hazy @Azaliamirh !

Introducing Contrastive Language Model (CLM): an ultra-fast System One Model trained with a contrastive learning objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and delivers up to 9× faster inference than Jev ⚡ while achieving comparable performance across computer-use, gaming, and tool-calling tasks. With lightweight fine-tuning, CLM-8B sets a new SOTA on challenging agentic coding benchmarks, such as DeepSWE (81.6%) and Terminal-Bench 2.1 (87.6%). In contrast, Jev fails to serve as an effective verifier for these long-horizon tasks. We also build an efficient training and serving infra for CLMs by disaggregating states and actions, allowing their embeddings to be cached and reused independently. This substantially reduces inference latency in settings where the state evolves continuously while the action set remains fixed. Finally, we establish scaling laws for CLMs and show that the test contrastive loss decreases predictably as a power law in training compute, model size, and dataset size. 📄 Blog: contrastive-lm.notion.site 💻 Code: github.com/Contrastive-LM/CL… 🗣️ Discord: discord.gg/5dAQEDJBs 🤗 Data & Models: huggingface.co/Contrastive-L… More details on CLM’s architecture, data recipe, and scaling laws in the thread below 🧵

I'm excited to announce @SnorkelAI's $350M Series E at $3.5B, led by @insightpartners and @S32_VC. We've grown 18x+ in the last 12 months since launching our Data-as-a-Service offering, passing $375M ARRR this week. As AI advances to superhuman capabilities, AI data & environment development must advance with it - and basic staffing and crowdsourcing approaches are not enough. AI progress now requires deep research and technology work that combines human expertise with specialized AI in compounding ways. @SnorkelAI is building the RSI data engine and frontier data lab for this next phase. We're honored to have the support of existing investors Addition, @lightspeedvp, @GreylockVC, @GVteam, P7, Factory, @WellsFargo, Walden Catalyst Ventures, and new investors @ThirdPointLLC, @MarchCPs, @BlumbergCapital, @AllegisCapital, @Frontlinevc, and @standard_vc. – @SnorkelAI started as a research project a decade ago at @StanfordAILab. Our thesis was simple: AI progress would become increasingly data-centric – and therefore data development should be studied as a true research and technology problem, not just a staffing and crowdsourcing one. Today, as AI capabilities verge on superhuman, building the data and environments to safely measure and train AI is becoming too hard for even the smartest human experts to do alone. Only humans and AI agents, collaborating together in compounding ways, can meet the accelerating needs of the frontier, and keep humans in the driver’s seat of AI progress for decades to come. At @SnorkelAI, we are building the data lab to define the shape of this new “Data 2.0” frontier, and the new paradigms of human-computer interaction needed to advance it. Our key focus is building the RSI engine for data, where specialized AI models accelerate and improve human expert output, and in turn, scaled human supervision is used to continuously evaluate and improve these models – creating a powerful compounding loop to keep pace with an accelerating RSI frontier. With this round of funding, we are also doubling down on our commitments to support data development for open benchmarking and evaluation (more news here soon!); an increasingly diverse ecosystem of general and specialized intelligence; and a path to safe, well-aligned AI built on robust training and evaluation data. Data development will guide and drive the next stages of AI – and must do so in a human-centric, AI accelerated, open, diverse, and safe way. We are excited to support this mission in the next decade of research ahead at @SnorkelAI. More thoughts here: snorkel.ai/blog/data-2-0-and…

Small models are truly capable of the majority of conversational use cases and even a majority of difficult reasoning ones. Amazing work @Avanika15 @JonSaadFalcon + team and great feature in @FT !

The inference landscape is going to get a lot more hybrid in the near future. We found that accuracy per joule of local models has improved 18x in just 16 months: 5.9x from hardware, 3.0x from model gains. Great in-depth cover by @FT: ft.trib.al/eVI82ZQ @Avanika15 @JonSaadFalcon John Hennessy @HazyResearch

dreams do come true 🥹. excited to see our work (w/@JonSaadFalcon, @HazyResearch, john hennessy and @Azaliamirh) feat. in a major way in @FT. the world is becoming increasingly less dependent on centralized cloud ai. we are just getting started 🚀🌖

Absolutely beautiful new paper from Stanford Univ + Together AI "Intelligence per Watt: Measuring Intelligence Efficiency of Local AI" - It found that hybrid local-cloud routing reduced energy, compute, and cost by 60% to 80% against its batched-cloud baseline. - From 2023 to 2025, local AI’s intelligence-per-watt improved 5.3×, while locally serviceable query coverage jumped from 23.2% to 71.3%. - An iPhone 16 Pro achieved roughly 7× higher intelligence-per-watt than workstation GPUs on the same model and precision, showing how power-efficient mobile AI can be for lightweight queries. - A diverse pool of 20+ local models actually beat the paper’s 3 frontier cloud models on 3 of 4 benchmarks when each query was routed to the best model, showing how much model diversity can matter. - Dropping precision from FP16 to FP4 cut inference energy by 3X–3.5X, while costing roughly 2.5 percentage points of accuracy per precision step; in one test, a larger FP4 model even beat a smaller FP16 model. - The biggest remaining weakness is concentrated at the hard end: on the paper’s hardest reasoning slice, about 95% of problems were still unsolved by local models, even as easier and medium-difficulty tasks improved rapidly.

impressive results from @Avanika15 and @JonSaadFalcon pushing the frontier of hybrid inference systems!

Some really cool speakers on harnesses for LLMs. How do we turn raw intelligence into an actual product? And how to engineering the software to scaffold AI to that end? Listen to @JonSaadFalcon and other speakers' on this.

we've built a local-cloud harness that is 800x cheaper & 4x faster than cloud-only harnesses. @JonSaadFalcon gives the tl;dr at the @ycombinator paper club. ff to 37:30 for more deets 🙂. link to full paper in comments 👇

Harnesses often get dismissed as just scaffolding, just prompt engineering, and not real research. But that couldn't be farther from the truth. The same model weights that score 30% on ARC-AGI score 95% with a better harness. So we gathered a group of researchers and founders working at the frontier to do a deep dive into the state of harnesses. We cover how we got to this point, the case for making your harness as expressive as possible, and what YC learned building an agent for every employee in the company. 00:00 - @FrancoisChauba1: Why harnesses matter 04:27 - Building an auto-researcher by accident 07:13 - A five minute history of harnesses 13:56 - Self-improving harnesses 18:35 - @sethkarten: Prime Agent, a self-improving RLM harness 21:50 - Context as an L1, L2, L3 cache 24:51 - From Turing machine to von Neumann computer 28:33 - Messaging between agents 30:04 - ARC-AGI results 33:09 - Emulator Bench and GPU kernels 37:30 - @JonSaadFalcon: OpenJarvis, personal AI on personal devices 38:26 - How far behind are local models 39:21 - The five primitives of a personal AI stack 42:47 - Letting cloud models optimize your local stack 43:53 - 800x cheaper than the cloud 45:58 - @josh__france and @jbellregan: QM, YC's agent harness for work 47:29 - A history of YC's internal agents 49:24 - OpenClaw and a fleet of 50 agents 51:04 - Pulling the brain out of the sandbox 54:43 - Letting the agent choose its own sandbox and model 57:16 - The grind tool: budgets on goals 58:50 - Agents don't understand social context

a great historical lens on the intelligence revolution from the 👑 @Konstantine . lots of gems on the implications of intelligence per watt improvements and the importance of gross domestic intelligence 🙌🏽

lfg @JonSaadFalcon. we live in an increasingly hybrid world 🙌🏽

The future of AI isn’t only in the data center. @JonSaadFalcon, joint Computer Science Ph.D. and MBA student at @Stanford, shares why he expects to see more language models built specifically for local use cases and what that shift means for the hardware and inference stacks behind them.

Local AI models can already answer 89% of everyday chat & reasoning queries as well as a frontier cloud model, a result that holds broadly across more than a million real queries & 20+ local models tested.