Tag: llm
14 articles
· ai / claude
Claude Sonnet 4 for developers — what changed from Claude 3
Sonnet 4 is a reliability upgrade for agentic work, not a raw benchmark jump. What changed in the API, where reward hacking dropped 69%, and whether to upgrade now.
· ai-tools / llm
Context engineering in 2026 — six patterns that work
Context engineering decides what your model sees at inference. Six patterns with code: ordering, caching, compaction, sub-agent isolation, and more.
· cloudflare / cloudflare-workers
How to Set Up Cloudflare Workers AI: Step-by-Step Guide
Run inference at the edge with Workers AI: scaffold a Worker, bind the AI, call models, stream SSE, generate images. Includes pricing and rate limits.
· nextjs / llm
How to stream LLM responses in Next.js with the Vercel AI SDK
Stream LLM responses token-by-token in Next.js using AI SDK v6. Covers route handler, client hook, and the two Vercel timeout traps most tutorials skip.
· llm / api
GitHub Models 2026 — free LLM API for developers reviewed
We tested GitHub Models' free-tier LLM API: rate limits, OpenAI compatibility, and whether 150 calls a day is enough for a real side project.
· ai-tools / llm
Prompt caching in 2026 — Anthropic, OpenAI, and Gemini compared
Prompt caching cuts costs 90%. Anthropic requires explicit markers, OpenAI caches automatically, Gemini bills hourly. Here is which one fits your workload.
· llm / openai
LLM structured outputs: JSON mode, function calling, and Zod
Grammar-constrained sampling is the only reliable LLM primitive. How OpenAI, Anthropic, Zod, and Vercel AI SDK v6 compare — and where each still fails you.
· claude / anthropic
Claude API 2026: Prompt Caching, Tool Use & Batches
A practical guide to the three Claude API features that separate toy prototypes from production integrations: prompt caching, tool use, and Message Batches API.
· openrouter / llm
OpenRouter vs direct API — when the gateway pays off
OpenRouter wins for multi-model projects and automatic failover. Direct API wins at high volume or for compliance-critical workloads. Here is how to decide.
· editors / zed
Zed AI in 2026 — how the built-in LLM features stack up
Zed AI is fast and private but lacks codebase indexing — behind Cursor on unfamiliar repos. Worth it if editor speed and BYOK matter more than semantic search.
· prompt-engineering / llm
Best prompt engineering tools for LLM apps in 2026
PromptLayer for PM-owned prompts, LangSmith for LangChain stacks, Braintrust for eval-first teams. Persona-grouped breakdown of 8 LLM tools, 2026.
· llm / cost-optimization
LLM cost routing: when Haiku beats Opus and when it does not
Routing 1M classification tokens from Opus 4.7 to Haiku 4.5 saves $6.00 — 80% reduction. Here is the task taxonomy, the latency case, and the tools to implement it.
· llm / fine-tuning
How to fine-tune a small LLM in 2026 (LoRA on a laptop)
Fine-tune Llama 3.1 8B with QLoRA on a consumer GPU — pinned Unsloth install, exact training config, GGUF export to Ollama, and eight failure modes.
· ollama / lm-studio
Ollama vs LM Studio on Mac — which survives daily use?
LM Studio wins on throughput and memory. Ollama wins on time-to-first-token and CLI setup. Here is when each choice makes sense on Apple Silicon.