Did TokenMaxxing survive post GTC?

Model routing is the present and future of production AI, driven by: - Cost - Skill - Availability

Nope. Prisoner’s dilemma proves that if any actor in a transaction can achieve leverage in a transaction by simply spending more on tokens (at any cost) to gain an advantage, they will

Well, it was good while it lasted I guess…

GLM 5.2 dropped at the right time

Intelligence should be open, accessible, and ready to build with, empowering every developer, everywhere. GLM-5.2 is now available to all GLM Coding Plan users, including Lite, Pro, Max, and Team plans. docs.z.ai/devpack/latest-mod… As our new flagship model, GLM-5.2 delivers powerful coding capabilities, usable 1M-context support, and continued strengths in long-horizon tasks. API and Chatbot services will launch next week. The model will also be officially open-sourced next week under the MIT License. The future of AI is open, and it belongs to the people.

OpenAI now just needs to sandbag the next model release So the US doesn't export control them So they gain massive market share It's imperative that all the OAI employees don't vauge hype post their next model release as the greatest thing ever Don't know if they are capable tho

Glad to see day-0 speculators are warm welcomed by the community!

MiniMax M3 weights are live. #1 open-weights model on the AA Intelligence Index.

🎉 Congrats to @MiniMax_AI on releasing MiniMax M3! Frontier coding and agentic capabilities, native image and video input, computer use, and a 1M-token context window, all in a single open model. At the heart of M3 is MSA, a new sparse attention architecture: instead of attending densely over the full KV cache, each query scores 128-token KV blocks and runs attention only over the top blocks. That is what makes 1M-token context practical to serve. M3 runs in vLLM with day-0 support, verified on NVIDIA and AMD hardware: ✨ MSA sparse attention with dedicated prefill and decode kernels ✨ 1M-token context serving with prefix caching and chunked prefill ✨ BF16 and MXFP8 checkpoints, with MoE backends for both Hopper and Blackwell ✨ Native multimodal input (image video) ✨ Tool calling, reasoning parsing, and thinking-mode control for agent workloads Day-0 support like this is a true team effort. Grateful to the teams at @MiniMax_AI, @NVIDIAAI, @AIatAMD, and @inferact, and to the vLLM community for making it happen. 🙏 Deep dive into the implementation, kernel work, and deployment recipes: 🔗 vllm.ai/blog/2026-06-12-mini…

I give it a year until we see a new breed of AI native private equity firms that acquire companies just so they can move their workflows from Claude to open source Chinese models and flip them.

Citadel Securities just put institutional weight behind what the AI bulls won't say out loud. In a new macro note titled "Tokenomics," Citadel makes the argument plainly: even the most powerful technology on earth still has to pass through the boring discipline of cost curves, capacity limits, and marginal returns. The evidence is piling up: – Amazon removed its token usage leaderboard – Microsoft cancelled Claude Code subscriptions – Multiple companies reporting unexpectedly massive token bills Their conclusion is the part that matters. Adoption is no longer about what AI can do in principle. It's becoming about the price and scarcity of the inputs needed to run it at scale. Compute. Power. Cooling. Memory bandwidth. Inference budgets. All real, all binding constraints. And here's the kicker from the chart. The Silicon Data LLM Token Expenditure Index, a benchmark for how much the market is actually spending on AI tokens, has started rolling over. Citadel reads it as a shift toward cheaper models. Companies substituting away from expensive frontier AI toward "good enough" alternatives. That's economics 101 doing what it always does. When the price of something rises, people use less of it, or find a cheaper version. Citadel sees a bifurcation forming. Frontier AI concentrated among a few firms with the balance sheets to absorb the cost. Everyone else quietly downgrading to simpler, cheaper models. This is the part of every technology revolution the early narrative ignores. The technology being real was never the question. The question was always whether the economics could carry the valuations. When one of the most sophisticated trading firms on earth starts writing about AI in the language of cost curves and rationing instead of limitless demand, the conversation has quietly changed. The hype was about what AI could do. The reckoning is about what it costs.

Congrats to @GoogleDeepMind on DiffusionGemma 🎉 A 26B diffusion language model on the Gemma4 backbone, and the first dLLM natively supported in vLLM. It denoises 256-token blocks in parallel instead of generating one token at a time: 1200 output tok/s at batch size 1 on a single H200 (FP8). Built on model runner v2's ModelState plus the existing speculative decoding path, with minimal scheduler or runner changes. FP8 and NVFP4 checkpoints are on the @RedHat_AI hub. Thanks to the @GoogleDeepMind, @RedHat_AI, and @NVIDIAAI teams! 🔗 vllm.ai/blog/2026-06-10-diff…

Subscription plans are massively subsidized. And by massively, I mean absurdly: Claude Max 20x: $200/month, with usage reportedly worth around $8,000 ChatGPT Pro 20x: $200/month, with usage reportedly worth around $14,000

Recently, we purchased one of each Anthropic/OpenAI subscription plan and randomly ran long horizon coding tasks until we exhausted the weekly limit. It's widely believed that a $200/month plan maxes out at ~$2000/month worth of tokens (assuming API pricing). However, we found that the subscriptions are actually far more generous. (2/4)

🎉 Excited to see Inferoa from @agenticin. It builds a community agent harness on the vLLM stack, with the agent loop shaped by inference economics: prefix-cache discipline, context optimization, and routing across self-hosted and frontier models. Looking forward to seeing how developers extend it. 🚀

THE AGE OF THE ROCKSTAR CFO I usually don’t post takes. But in today’s newsletter I wrote a take about why the CFO will be seen as more strategic than ever as companies face new and interesting questions on both capex and opex

Super excited to announce that @arcee_ai is the first major American AI lab to replace AWS S3 with Hugging Face for ALL their models and datasets, public AND private 🔥🔥🔥 Multi-million $ partnership to support American open-source AI, let’s go!

Want to work on a datacenter with 1k GPUs? I'm hiring people to write code for our small 500kw cluster (~3MW in couple years). You'll work with things like SLURM, PyTorch, NFS, Linux, miniray, minikeyvalue.

If you've adopted AI at your company but haven't seen any tangible results, read this 1990 article: "The Dynamo and the Computer" by Paul David. When electricity first arrived, factories that "adopted" it barely got faster. They just swapped the steam engine for an electric one and ran everything else exactly as before: same machine layout, same workflow, same management. Electricity in, no real gains out. The most common mistake with any new technology is to drop it into the old organization and then declare the transformation done. The real leap came decades later, when each machine got its own small motor. Suddenly machines no longer had to be lined up around one central drive shaft. They could be rearranged around the actual flow of work. The productivity gains didn't come from electricity. They came from REDESIGNING THE ENTIRE FACTORY around it. AI is the same. Bolting it onto your existing process gets you a faster steam engine. The payoff comes when you redesign the work itself. (link to paper in comments)

Interesting approach from Apple They are storing the shared attention block in the DRAM While the FFN weights stay in NAND and are loaded in the DRAM, depending on the request Apple is facing 3 constraints - 1) Limited DRAM size 2) Large model size (20B params) 3) Slow NAND read speed A super small model (sub 8B) won't be that useful, but they can't store a 20B model in DRAM (due to memory shortage). They also have to manage the KV cache overhead. If they streamed the weights completely through iPhone SSD, then it would take 2.5 seconds to generate just 1 token (0.4 tokens/s) So the big thing here is that a normal MoE activates different experts based on every token, but in Apple's case, a sparse mask predictor decides which parameters to activate based on the request/prompt, locks it in, and loads it into the DRAM (1B-4B depending on the request). They basically convert a 20B MoE (with 1B-4B active) into a dense 1B-4B param model for a request. The tradeoff: They are basically adding 0.3-1.5 seconds (1B to 4B params loaded) of latency to TTFT time by loading FFN weights from NAND to SSD per request (read speed is around 1.5-1.7 GB/s for iPhones) and taking a hit to performance They will get around 15-50 tokens/s of decode speed (depending on params loaded) Ideally, smartphones would come with 24-32 GB of RAM so that 20B param models could be loaded, but memory shortage won't allow it to happen But, their competitor here is ChatGPT Instant, which is a much smarter model that runs at 200 tokens/s and has a TTFT of 0.8 seconds (Apple's TTFT will be around 0.5-2 seconds, and decode speed is around 15-50 tokens/s), and is also free Apple's AFM on device models will be great for privacy-focused tasks. They get beaten by cloud models on other benchmarks (perf, speed, quality)