Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me.
I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA.
I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation.
Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO.
Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours.
Jensen
@mttjon This comment is actually nothing short of ignorance. Conflating specific stable coin authorities with decentralisation. Blockchain isn’t this magic decentralised ecosystem. There exist many, many centralised systems within blockchain and a CEX using stable coins is one of them.
I share the same sentiment. Why should there be so much friction to access these latest models? I’m not even remotely interested in jumping through all these hoops to access a model that may or may not even be marginally better than the competition
i'm not testing muse wonder, because:
1. without contributor tier, it's opus 5.5 at API price. why don't i just run opus 5.5 then...
2. it's weird to lock the "full" model behind their subscription
i'm also not testing gemini 4 argon, because:
1. previous gemini models had lost my trust, and nothing so far gives me enough motivation to revisit
2. google's AI subscription doesn't allow the models to be used in 3p harnesses
i know this sounds harsh to those companies, but i see many consumers are in a similar situation - we have way too many options, some of which are absolutely AMAZING right now. the bar is extremely high for us to justify spending any second of our time on an unproven model family
meta and google need to both realize they are catching up on market perception, and are in no position to gate anything
my suggestion to them would be to allow heavily subsidized access everywhere including openrouter and all 3p harnesses, and leverage that to get enough people to try their models. if the models are good, the words will then spread
@theo Do you plan on decoupling repos from projects? I would like to start a session from inside a parent folder and have it delegate tasks to my child repos. This functionality feels a bit sticky right now in T3Code
Hey Daniel, big fan of your infosec work.
However, don’t you think this is slight information overload? How does this data assist with decision making, and why is it best placed inside the claude code instance. I assume your workflow does not include terminal multiplexers? Would be a headache reading it all with the verbose status line text
Aloha! 🌺Introducing Ornith-1.5, a family of open-source LLMs spanning 9B Dense, 35B MoE, and 397B MoE, trained with self-improving strategies.
It achieves state-of-the-art performance among open-source models of comparable size and delivers performance comparable to Claude Opus 4.8 across reasoning, agentic, and coding tasks:
✅Terminal-Bench 2.1 (86.1)
✅SWE-Bench (86 on verified, 65.1 on pro, 79.6 on Multilingual)
✅DeepSWE (56)
✅HLE (44.6)
✅ClawEval (81.4)
✅Tool Decathlon (71.2)
Ornith-1.5 takes a major step toward training foundation models through end-to-end self-improvement, extending the self-scaffolding strategies introduced in Ornith-1.0 into a more complete self-improvement loop: the model proposes new tasks, generates task-specific scaffolds, and produces solution rollouts for reinforcement learning, continuously creating new learning experiences from which it can improve.
All models, along with their quantized versions (FP8, GGUF, MLX, and NVFP4), have been released under the MIT License, enabling unrestricted commercial and research use.
📘Tech Blog: https://t.co/OZ63scRWLB
🤗Huggingface: https://t.co/mGJLwhrQOM
So, the notion of parameter maxxing is essentially obsolete
- The trillion-parameter era was a learning experience. The 2020 scaling law (grow params 2.7x faster than data) was refit in 2022, the real optimum is ~20 tokens/param
- Inference now dominates lifetime cost, so the meta flipped to small models massively over-trained e.g. Gemma-2-9B saw ~889 tokens per parameter. Chinchilla-optimal no longer applies
- MoE models, more total parameters can actively hurt reasoning. Total params = memory. Activated params = thinking. At fixed data ratios, adding knowledge degrades the ability to reason while activating more experts reliably helps
- The capability gains are no longer gated on pretraining a bigger base. GLM-5.3 is the same model as 5.2, same params, same architecture plus one month of post-training/RL
Essentially the core takeaway is that the models of the future won't just be bigger, they'll be trained differently. Welcome, the post-training era
Thoughts About Scaling Law
Scaling, but not only of parameters. Every model release now ends with the same question: how many parameters? It isn't a question that can be answered on its own. Parameter count is only meaningful alongside three others — how much data you have, where you intend to spend your compute, and who will run the model, under what conditions.
The field learned this the hard way. Kaplan et al. (2020) fit an exponent that told everyone to grow parameters faster than data — roughly 2.7:1 — and the industry complied: GPT-3, Gopher, MT-NLG. Hoffmann et al. (2022) redid the experiment across four hundred models and found the compute-optimal split is closer to 20 tokens per parameter, and that with sufficient compute the two should grow at the same rate rather than drifting apart. The error in the earlier fit compounded with every order of magnitude of compute, which is why the largest models of that generation were the most misallocated. The trillion-parameter round was, in retrospect, a detour the whole field took together and then reversed.
Chinchilla wasn't the end either. It optimized training compute for models that would be trained once and evaluated. Today a model is called billions of times a day and inference dominates lifetime cost. Put inference into the objective and the optimum moves toward smaller models trained far longer — deliberate over-training, which is what Llama-2-7B and Gemma-2-9B were doing at roughly 290 and 889 tokens per parameter.
Sparsity moved the target again. In a MoE model two quantities have to be kept apart: total parameters govern roughly how much the model can hold — knowledge, facts, the long tail — while activated parameters and effective depth govern roughly how far it can think, how many steps of a causal chain it can carry before it comes apart. A dense 20:1 ratio does not transfer. And the ratio isn't a single number at all: Roberts et al. (2025) find the optimal tokens-per-parameter is task-dependent, with memorization favoring more parameters and reasoning favoring more data. Follow-up work on MoE observes that at fixed TPP, pushing total parameters higher actually degrades reasoning, while activating more experts reliably helps it.
This matters for what we are building toward. Finding a vulnerability is not a retrieval problem. It doesn't come from having memorized more CVEs; it comes from carrying a twenty-step chain of inference to the end without losing the thread. That capability does not live in total parameter count.
Which brings us to this release. Total parameters appear to matter up to a threshold — enough to hold the world — after which additional capability comes from scaling elsewhere: effective depth per forward pass, and above all post-training. GLM-5.3 is our controlled experiment on that claim. Same base, same architecture, same total and activated parameters as GLM-5.2. One month of scaling long-horizon environments and RL. The gains are not marginal. Well, scaling has more than one dial. We turned the post-training one this time because it had the most slack left in it — not because the others are finished. Base model size, pretraining data, compute spent per forward pass: all of them are still on the table, and we will come back to each. What this experiment taught us is that the dials do not have to be turned together, and that the one worth turning next is rarely the one that was worth turning last. We are not done scaling. Next time, maybe mid-training, pre-training, and even more.
Keep up the great work @liquidai - the community appreciates these model sizes.
Just released the uncensored version of LFM2.5-2.6B
https://t.co/CLycnkW6a2
Today we release LFM2.5-2.6B, an agentic model that runs entirely on-device. It plans, calls tools, and works through multi-step tasks on phones, laptops, PCs, and robots. Data never leaves the device, and the marginal cost of each run is essentially zero.
> Pre-trained on ~34T tokens
> LFM2.5 flagship hybrid architecture
> Context length: 128K
> Vocab size: 128K
> balanced intelligence per watt
> customizable on a single GPU for any specialized task
> LFM2 open-weight license
Comparable or better scores compared to models up to nearly 4x its size:
> ToolSandbox 77.83, ahead of Qwen3.5-9B at 76.44
> Multi-IF 80.07, ahead of Gemma-4-E4B-it at 77.35
> IFStruct 85.49, ahead of Qwen3.5-9B at 78.50
🧵
@alexcooldev I have found I can more reliably run long horizon tasks with Kimi. In particular end to end testing of my IOS apps overnight. With Opus I have found some regressions cropping up during end to end testing - it does correct itself, Kimi just completes faster