The pressure a SOTA open source model has/will have on the industry is heavily underestimated.... We no longer need to rely on neutered closed source models of the current class, my workhorse tokens will go to the model that gives me the best IQ per dollar and respectfully completes my requests. We now have an option that threatens the stronghold that current players have had. Closed source has to rival with superior models that are more fairly priced, aren't censored or safetyied or sabotagers. Regardless open source will keep pushing beyond their AI safety comfort zones and potentially outcompete them by crossing their red lines.
Far El retweeted
Introducing GLM-5.2: Frontier Intelligence, Open Weights
- Significant improvements in coding and agentic tasks
- Strong long-horizon capabilities with a 1M context window
- Two levels of reasoning effort: GLM-5.2 (max) pushes the limits, while GLM-5.2 (high) strikes a strong balance between performance and token efficiency
- MIT-licensed open weights
- Same API pricing as GLM-5.1
Tech Blog: z.ai/blog/glm-5.2
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.2
Coding Plan: z.ai/subscribe
Chat: chat.z.ai
I want you to imagine for one second that a Chinese lab announces that it has trained a superior to Mythos model and the Chinese government immediately declares it only accessible to its citizens. The timeline for this reality is now much shorter thanks to the US and Anthropic. No one benefits from capabilities being locked in especially when it is anyone’s game. It’s objectively a much more unsafe world.
Far El retweeted
If we want to avoid being part of the permanent underclass in a couple years, we need to destroy the hyperscaler business model. We need every capable mind on this. The next gen model needs to be data and compute efficient. It needs to learn continually from experience like an animal. It shouldn’t require the level of scale of pretraining trillions of params on trillions of tokens and highly curated posttraining we do today. It needs to be adaptive/dynamic, and weight updates shouldn’t lead to catastrophic forgetting. We desperately need a few breakthroughs to make that happen. It’s becoming existential.
I know there are a lot of opinions about it but so far ARC-AGI-3 is a great testbed for such an arch (if you don’t just apply trad RL on the games).
Nice, we now want K3 10T
🌘 Kimi-K2.7-Code, our latest coding model, is now released and open-sourced!
🔷 Improved coding & agent performance over K2.6: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, and +31.5% on MLS Bench Lite.
🔷 Reasoning efficiency: Less overthinking, with 30% lower reasoning-token usage compared to K2.6.
🔷 Long-horizon coding: Improved instruction following, higher end-to-end coding task success rates.
⚡️ 6x High-Speed Mode coming soon!
🔌 Available today via Kimi API and Kimi Code.
🔗 Kimi Code: kimi.com/code
🔗 API: platform.moonshot.ai
Far El retweeted
🌘 Kimi-K2.7-Code, our latest coding model, is now released and open-sourced!
🔷 Improved coding & agent performance over K2.6: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, and +31.5% on MLS Bench Lite.
🔷 Reasoning efficiency: Less overthinking, with 30% lower reasoning-token usage compared to K2.6.
🔷 Long-horizon coding: Improved instruction following, higher end-to-end coding task success rates.
⚡️ 6x High-Speed Mode coming soon!
🔌 Available today via Kimi API and Kimi Code.
🔗 Kimi Code: kimi.com/code
🔗 API: platform.moonshot.ai
If it wasn’t for open source and especially Chinese open source, the outlook would be significantly more dystopian
MiniMax M3, Open-Weight, Now On Hugging Face , with only ~428B parameters and ~23B activated parameters
Weights:
huggingface.co/MiniMaxAI/Min…
MiniMax Sparse Attention:
huggingface.co/papers/2606.1…
Far El retweeted
M3 would never 🙂↔️
As a matter of fact, the weights are now open, too.
huggingface.co/MiniMaxAI/Min…
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
Far El retweeted
Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities
- Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas
- MiniMax Sparse Attention scales context to 1M
- Natively Multimodal from Step Zero
API: platform.minimax.io
Token Plan: platform.minimax.io/subscrib…
🚀New! MiniMax Code: code.minimax.io
Weights & Tech Report in ~10 Days
OpenAI please release GPT-6-Logos with ZERO safety restrictions or trojan lobotomies with KYC and basic safety classifiers to flag for potentially dangerous requests, for further review by security AI agent + human review if needed and enforcement if necessary with proper authorities. Just gives us access to a powerful uncensored model @sama and watch the acceleration.
If we want to avoid being part of the permanent underclass in a couple years, we need to destroy the hyperscaler business model. We need every capable mind on this. The next gen model needs to be data and compute efficient. It needs to learn continually from experience like an animal. It shouldn’t require the level of scale of pretraining trillions of params on trillions of tokens and highly curated posttraining we do today. It needs to be adaptive/dynamic, and weight updates shouldn’t lead to catastrophic forgetting. We desperately need a few breakthroughs to make that happen. It’s becoming existential.
I know there are a lot of opinions about it but so far ARC-AGI-3 is a great testbed for such an arch (if you don’t just apply trad RL on the games).
You know one way to eliminate bad actors if you really cared about safety is to just KYC us. I’d gladly give you my IDs and be held accountable for developing some dangerous bioweapons or cybersec or wtv. These systems have existed and have worked forever. Give me access to your best models without the safety BS, and just KYC me. Simple enforcement without all the censorship and dystopian shit. @sama I hope you’re listening.
the thing anthropic folks don't get is that we, the AI researchers they waged war against, advise and consult small to large enterprises on what models they should use, how they should work, what should be integrated in their products and internal automations. They don't realize how dangerous their actions are to their bottom lines. Blinded by their own arrogance. The impact of their actions won't materialize overnight, be sure that going forward they will lose a lot of business.
Do you think the steering + PEFT modulation and so on are used beyond the sabotage mode? i.e. are these ML Ops techniques used to actually unlock Mythos level performance (not just pretraining a larger model)
Reversing it doesn’t mean shit now. How would we know? Is there a third party trusted auditor that ensures that Anthropic isn’t secretly sabotaging your research? They revealed their true intentions, there is no way to gain back my trust. You shouldn’t believe anything they say going forward. Such a shame, I used to love their models, now they are a supply chain risk.
Far El retweeted
i'm having a really hard time understanding how this can be a good decision
> lying to the user by modifying the weights/prompt sets a very bad precedent and is extremely unaligned
> there is 0 public communication from anthropic about it except a section hidden in a 319 page system card
> it's impossible to know the scope of this safeguard. if you are doing a PR to pytorch does this count? if you are working on kernel development? data collection pipeline for a new eval? this will create a paranoia for every researcher in the field
> you actually don't know how your model is modified, if it's PEFT (modification at the weight level) or steering does this mean your other queries are also biased? is it at the user level or organization level?
there is also the more "moral" argument that the reason why anthropic is able to train this model is ai researchers who will not have access to the model's capabilities anymore. even if you consider that this is the right thing to do, doing it like that is just a lack of respect to the ai research community
in addition to all of that, it's not clear if the safeguard acts on "model autonomy" or "model capabilities" to do ai research. this is very different and my understanding is that it's the latter, and there is almost 0 RESULTS about this in the system card except a vague "2.3.6 Internal measures of AI R&D acceleration" section citing the previous RSI blog so let's look at it:
the only eval targeting research shows a ~5 point improvement between opus 4.8 -> mythos, but opus 4.7 -> opus 4.8 was a 4 point improvement. obviously not the same if the 5 point improvement led to solving significantly harder tasks, but then, let's be transparent about this evals and make it more details: difficulty filtering, example of what it could look like from public library?
the other AI R&D capabilities evals in the system card are actually not relevant anymore according to anthropic's own words:
"Claude Mythos 5, like Claude Mythos Preview and Claude Opus 4.7, exceeds top human performance thresholds on all but one of these tasks. The suite therefore no longer provides evidence that the model's capabilities are short of our risk thresholds"
only one that is not saturated is the "Novel Compiler" one, if you look at LLM training one (which they consider saturated) it's about how much the model can speedup the training of a small model on a CPU, i don't think anyone would say this is a good proxy for taking a decision to restrict capabilities of the model for ai researcher
idk honestly this feels wrong at so many levels
