Nearly every frontier open-source model in 2026 is MoE: Ornith 35B activates only 3B to beat 31B dense models, MiniMax M3 uses 456B total but 45.9B active to hit SWE-bench Pro 59%, DeepSeek V4 runs 1.6T total with 49B active. This post explains why MoE dominates coding and agentic benchmarks using four case studies.
Models don't read words — they read tokens. A Chinese character is typically 1-2 tokens; an English word is 1-3. The context window is the token limit per request. Inference is using a model; training is teaching one. What you do every day is inference.
A 70B model needs ~140GB VRAM in FP16, but 4-bit quantization shrinks it to ~35GB. With llama.cpp's partial CPU offloading, it can run on consumer hardware. GGUF naming conventions (Q4_K_M, Q5_K_S) tell you the precision-size tradeoff. KV cache is why long conversations slow down.
Text Generation Inference (TGI) is HuggingFace's own LLM inference server, built in Rust and Python. It pioneered continuous batching and Flash Attention in open-source inference engines. The GitHub repository was archived on March 21, 2026, and HuggingFace recommends migrating to vLLM or SGLang. TGI still matters: it defined the architectural baseline that successor engines inherited, and many HuggingFace Inference Endpoints still run it.
Cerebras can dramatically accelerate generation on supported models, but agent latency still depends on prefill, tool I/O, model quality, and platform compatibility.
Lecture 13 moves from inference efficiency to inference capability: speculative decoding drafts with a small model and verifies with a large one; on-policy distillation addresses drift; long context and test-time scaling spend inference resources.
Fireworks AI puts open-weight model evaluation, dedicated GPU deployments, and LoRA customization behind one API surface. Serverless fits low-volume starts, On-demand fits sustained traffic and custom models, while reserved capacity adds enterprise capacity guarantees.
Hugging Face Hub is a collaboration layer for versioned models, datasets, and applications. Datasets handles data, Spaces runs demos, while Inference Providers and Endpoints provide managed inference.
Together AI puts serverless APIs for open-weight models, dedicated GPU endpoints, batch inference, and fine-tuning on one platform, letting teams validate per token before moving to reserved deployment when traffic or customization justifies it.
CoreWeave is more than rented GPUs: it combines Kubernetes, GPU networking, storage, and inference into AI infrastructure, while platform engineering and capacity governance remain yours.
Nebius combines GPU VMs and clusters, Kubernetes, Slurm, storage, and Serverless AI; choose the responsibility layer before comparing hardware and price.
Ray Serve is a distributed serving layer on Ray. Deployments and handles compose Python service graphs, while replicas, CPU/GPU scheduling, autoscaling, and model multiplexing handle orchestration; it complements rather than replaces vLLM or SGLang.
Replicate abstracts GPUs behind versioned models, predictions, Cog, and deployments; integrators still own version pinning, async workflows, webhook verification, data persistence, and spending limits.
RunPod Pods fit interactive and persistent GPU work, while Serverless fits queued or load-balanced inference; choosing incorrectly mixes persistence, cold starts, and retry semantics.
SGLang is an inference engine for generative models. RadixAttention reuses KV cache across shared prefixes, while OpenAI-compatible APIs, structured output, and multi-GPU parallelism support production LLM serving; it is not a complete product backend.
Triton Inference Server serves TensorRT, ONNX, PyTorch, and other models through consistent HTTP and gRPC APIs. Its defining tools are the model repository, dynamic batching, instance groups, and ensembles—not LLM-specific KV-cache scheduling.
Google PMLE, AWS AIF-C01 and AIP-C01, Microsoft AI-103 and AI-500, and NVIDIA NCP-GENL all test how to make a GenAI application fast, cheap, and reliable — and they form a three-rung ladder: AIF-C01 asks whether you know cost scales with tokens, AIP-C01 and the two Microsoft exams ask whether you can instrument and control it, NCP-GENL asks whether you can change the model and the hardware. Three different altitudes. NVIDIA works at the kernel and quantization layer (Model Optimization 17% + GPU Acceleration 14% = 31%, the heaviest single cost/latency block in the whole series), AWS and Microsoft at the application layer (three caching tiers, token caps, chargeback), and Google at the MLOps layer (CPU/GPU/TPU evaluation, data vs model parallelism, scaling serving backends by throughput). The shared core is eight levers, but each lever becomes a different question at each altitude. This post deliberately carries no prices and no hardware specs — that is the part of this topic that rots fastest.
For side projects, toy demos, and RAG prototypes, nobody wants to swipe a credit card on day one. This is a verified roundup of 40+ LLM inference providers still operating as of 2026/05, tiered by whether free resources auto-replenish or are one-time grants. Each entry notes credit-card requirements, supported models, paid starting prices, and catches. Chinese-origin providers including Zhipu GLM (permanently free), Doubao (2M tokens/day), Kimi, DashScope, and the Ollama local option are all included.
Groq Console is the developer portal for Groq's in-house LPU chip, offering an OpenAI-compatible API, Playground, and free tier credits. Its selling point is running open-source models like Llama, Qwen, and DeepSeek at the fastest tokens/second on the market.