Mixture-of-Experts
6 stories filed under Mixture-of-Experts on Beyond Market Intelligence. The newest of them: “Explore how Kimi K3 makes powerful AI more accessible and affordable.”, “Enterprise AI agents gain smarter control as token costs drop by 52 percent”, and “Nvidia's Switchyard routes AI tasks to the right model, slashing costs”. A 2.8-trillion-parameter model that only activates a fraction of its weights per token sounds like a paradox, but that's exactly what makes Kimi K3 interesting. Enterprises watching AI agents multiply across their workflows have hit a wall: token bills that climb with every autonomous task. Acme AI is the next-generation, AI-powered spreadsheet platform built to replace Excel and redefine how analysts, data scientists, and enterprise teams work… The list below is every Mixture-of-Experts story on Beyond Market Intelligence, newest first.
Explore how Kimi K3 makes powerful AI more accessible and affordable.
A 2.8-trillion-parameter model that only activates a fraction of its weights per token sounds like a paradox, but that's exactly what makes Kimi K3 interesting. Moonshot AI built it with a Mixture-of-Experts architecture to keep inference costs down while pushing coding and agentic performance. It's a practical alternative to proprietary systems, especially when you weigh open weights against near-frontier capability. We think that trade-off is worth exploring.
Enterprise AI agents gain smarter control as token costs drop by 52 percent
Enterprises watching AI agents multiply across their workflows have hit a wall: token bills that climb with every autonomous task. Writer's new Palmyra X6 model attacks that problem head-on, cutting agent costs by 52% while improving speed by 48% and quality by 10%. The real news isn't just the numbers, it's the strategy. By post-training an open-weight base model on just 626 curated examples, Writer proves the enterprise AI race isn't about building from scratch.
Nvidia's Switchyard routes AI tasks to the right model, slashing costs
Nvidia's Switchyard router cuts agentic costs by matching each step to the right model, and paired with the new Nemotron 3.5 Lightning, it holds frontier-level quality at roughly a third of the price. That's the fix enterprises have been circling. The real bet is owning both layers, model and routing, under one open license. It shifts the competitive question from which model is best to which system routes best.
Inkling Small delivers near-flagship performance at a fraction of the size.
Just two weeks after unveiling Inkling, Thinking Machines is back with Inkling-Small, and it's a compelling trade. This 276-billion-parameter model nearly matches its 975-billion-parameter predecessor, scoring 40 versus 41 on the Artificial Analysis Intelligence Index while using roughly a third of the active parameters. It even outpaces the flagship on coding and reasoning benchmarks. For enterprises, that's a meaningful shift: near-flagship capability with lower compute demands and a permissive Apache 2.0 license. It's not laptop material, but it's a far more practical self-hosting option.
Tiered optimizer cuts MoE training memory demands by 97 percent
SkewAdam tackles the biggest pain point in MoE training head-on: optimizer state memory. By tiering precision across backbone, experts, and router, it cuts state from 50.6 GB to 1.29 GB, a 97.4% drop that lets a 6.78B MoE fit on a single 40GB GPU. That's not just incremental progress; it's a practical unlock for researchers stuck on smaller hardware. For deeper context on why memory efficiency matters, our coverage of Microsoft's TauGrid shows similar pressure points in AI workload management.
Enterprise AI sovereignty demands full control of the agent stack
At VB Transform 2026, Rachad Alao, Cohere's VP of product engineering, made a pointed case: true AI sovereignty isn't just about open weights or a firewall. It demands control over the entire stack, from GPUs to governance. That's a message that resonates when banks and hospitals are involved. Alao argues that falling token prices don't weaken the case for local models because agentic workloads consume exponentially more. It's a practical push for model routing and data control, not hype.