José David Baena – Software Engineer https://josedavidbaena.com Personal website and blog where I document thoughts, ideas, and interests in software engineering, web performance, and open source technologies. en-us josedab@gmail.com (José David Baena) josedab@gmail.com (José David Baena) Sat, 01 Aug 2026 00:00:00 GMT https://josedavidbaena.com/blog/kimi-k3/06-lab-experiment-roadmap Kimi K3 Ideas Need Experiments Before Design Rules https://josedavidbaena.com/blog/kimi-k3/06-lab-experiment-roadmap Turn Kimi K3 mechanisms into draft experiments with fixed controls, repetitions, uncertainty reporting, and explicit limits on every conclusion. Sat, 01 Aug 2026 00:00:00 GMT josedab@gmail.com (José David Baena) kimi-k3experimentsablationreproducibilitymodel-architecture https://josedavidbaena.com/blog/kimi-k3/05-api-vs-self-hosting-cost Kimi K3 Cost Model: Hosted API Versus Self-Hosting https://josedavidbaena.com/blog/kimi-k3/05-api-vs-self-hosting-cost Model Kimi K3 API and self-hosting costs with real token mixes, topology quotes, measured throughput, redundancy, storage, and engineering overhead. Fri, 31 Jul 2026 00:00:00 GMT josedab@gmail.com (José David Baena) kimi-k3api-pricingself-hostingcost-optimizationinference https://josedavidbaena.com/blog/kimi-k3/04-checkpoint-parameter-accounting How 2.8T Parameters Fit in a 1.56 TB Kimi K3 Checkpoint https://josedavidbaena.com/blog/kimi-k3/04-checkpoint-parameter-accounting Reconstruct how Kimi K3's 1.561 TB snapshot combines packed FP4 values, scales, BF16/F32 tensors, shard overhead, and repository files. Thu, 30 Jul 2026 00:00:00 GMT josedab@gmail.com (José David Baena) kimi-k3quantizationmxfp4model-weightscheckpoint https://josedavidbaena.com/blog/kimi-k3/03-api-production-state-tools-caching Kimi K3 API State: Tools, Caching, and Failure Modes https://josedavidbaena.com/blog/kimi-k3/03-api-production-state-tools-caching Kimi K3 is conversation-stateless at the API boundary; preserve full assistant messages, tool results, cache identity, and terminal-state checks. Wed, 29 Jul 2026 00:00:00 GMT josedab@gmail.com (José David Baena) kimi-k3apitool-callingcontext-cachingproduction https://josedavidbaena.com/blog/kimi-k3/02-architecture-kda-attnres-experts Inside Kimi K3: How KDA, AttnRes, and 896 Experts Work https://josedavidbaena.com/blog/kimi-k3/02-architecture-kda-attnres-experts Trace Kimi K3's 69 KDA layers, 24 gated MLA layers, Attention Residuals, and 896 experts without overstating what released artifacts prove. Tue, 28 Jul 2026 00:00:00 GMT josedab@gmail.com (José David Baena) kimi-k3architectureattentionmixture-of-expertsmachine-learning https://josedavidbaena.com/blog/kimi-k3/01-hardware-requirements Kimi K3 Hardware Requirements: Why 8 H100s Are Not Enough https://josedavidbaena.com/blog/kimi-k3/01-hardware-requirements Kimi K3 activates about 104.2B parameters per token, but its 1.561 TB snapshot still needs a multi-GPU memory, cache, storage, and network plan. Mon, 27 Jul 2026 00:00:00 GMT josedab@gmail.com (José David Baena) kimi-k3gpuinferencemixture-of-expertshardware