Topic
A neural network design that activates only a few specialized "expert" subnetworks per input, scaling parameter count without raising compute cost.
The Neuron measured NVIDIA's Nemotron 3.5 Bolt 74B-A9B at 65.5 tokens per second on a laptop in a hands-on test. The build was heavily compressed, and the review's tool-assisted wins went to the hosted Super VL.
Reality
Meta released Muse Spark Open, an 816B-parameter open-weight model that activates 42B parameters per token. Per The New Stack, a company above $20M in revenue needs a separate Meta agreement before selling it as a hosted service.
Perspective Coverage
13 publishersReality
Reflection AI's Beam coding model computes 23B parameters per token but keeps all 501B in memory, according to AWS and Hugging Face. Self-hosting it starts from a roughly 1 TB BF16 checkpoint and multi-GPU sharding.
Reality
NVIDIA says its Nemotron 3.5 Bolt 74B-A9B agent model fits in 24GB and will ship with open weights, data and recipes. The footprint, the license terms and the quality claims stay unverified until the weights are live and outside testers have run them.
Perspective Coverage
12 publishersReality
NVIDIA's open Nemotron 3.5 Super VL is a 120B mixture-of-experts model that activates 12B parameters per token for text, image, document and video work. NVIDIA's post still places it in a multi-model system, so consolidation covers one tier.
Reality
Liquid AI says its LFM2.5-24B-A2B scores 48.09 on LiveClaw against 38.19 for the larger Qwen3.6-35B-A3B. The 184 tok/s it reports on an Apple M5 Max is a decode rate, so agent latency depends on how many tokens the model spends thinking.
Perspective Coverage
4 publishersReality
Fish Audio released a 35B mixture-of-experts speech model on Oct 13 that scores 93.2% on Artificial Analysis' pronunciation benchmark. Free commercial use stops at $500,000 in annual revenue, and the two-RTX-3090 setup is one Reddit user's report.
Reality
Ai2's Olmo-core 3 reached about 52,000 tokens per second per GPU on a 47B mixture-of-experts model, roughly 2.7 times the old stack, on eight B300 GPUs. The report labels every headline rate as measured under random routing, so a team with a trained router has to rerun the comparison.
Reality
Janas, a C engine by Maurizio Cammalleri, reports 37 tok/s on a 48.4 GB model with 32 GB of RAM, against 2.8 on llama.cpp. Speculative drafting supplies 13 of those 37, so the like-for-like figure is 24 tok/s.
Reality
Qwen released eight open-weight Qwen3 models, two MoE and six dense, that each support thinking and non-thinking modes. On the model card, one chat-template argument, enable_thinking, picks the mode and defaults to True.
Reality
Mistral AI put Large 4, a 1.05-trillion-parameter mixture-of-experts model, into public preview on Oct. 6 and promised weights by October's end. Self-hosters can size memory now, though Mistral says the training run behind the preview is still going.
Reality
Aleph Alpha published Kolibri, a 78B-parameter English and German mixture-of-experts model, under Apache 2.0 with about 3.46B parameters active per token. Teams that cannot send prompts to outside servers can host it themselves if they pay for the GPUs.
Reality
Qualcomm CEO Cristiano Amon says AI firms want phones running 100-billion-parameter models continuously by 2028, more than triple his newest chips' limit. Most of Qualcomm's plan for closing that gap is memory, the cost Amon says already deters phone upgrades.
Reality
Xiaomi's $850,000 and $2.62 million cover 30 reinforcement-learning steps apiece on models it had already pretrained. Xiaomi published the environments and the RL code but not the 7,000-plus task datasets behind them.
Perspective Coverage
7 publishersReality
Artificial Analysis measured the score itself, and the weights are on Hugging Face under MIT, but MiMo-V2.6-Pro holds 1.02 trillion parameters, so most teams will reach it through an API at $0.43 per million input tokens.
Reality
The quarter-size global KV cache and eighth-size persistent storage are serving-cost claims rather than benchmark scores, and banking the second one requires a prefix-cache tier your stack has to already run.
Perspective Coverage
6 publishersReality
DeepSeek V4.1 Flash costs $2.10 off-peak for an agent run that costs $75 on Claude Opus 5, about 36x less, according to a dev.to analysis. Paying peak-hour rates halves that gap to about 18x, and running at max effort cuts it to about 15x.
Reality
Reflection AI says its first model, Beam, performs at the level of Z.ai's GLM-5.2 while using three to four times less compute than similar open models. Until outside testers report, those are the company's figures, and they count most with buyers who cannot use Chinese models.
Perspective Coverage
12 publishersReality
VIDRAFT released a 111GB, 4-bit build of its 180B-parameter Darwin model that it says runs on a laptop with 8GB of VRAM and 32GB of RAM. Air-gapped teams still need a measured token rate on that laptop before they can price it against a server.
Reality
Reflection AI says Beam, its 501-billion-parameter open-weight model, needs three to four times less compute than comparable open models. Buyers have only the company's figures until the weights ship later in October.
Perspective Coverage
13 publishersReality
Build · October 3, 2026 · Confirmed6 publishers
Build · October 2, 2026 · Confirmed2 publishers
Build · October 1, 2026 · Confirmed2 publishers
Build · August 20, 2026 · Confirmed2 publishers
Invest · August 30, 2026 · Confirmed2 publishers
Build · September 11, 2026 · Confirmed3 publishers
Build · September 20, 2026 · Confirmed2 publishers
Build · September 15, 2026 · One report1 publisher
Build · September 11, 2026 · Confirmed2 publishers