Tag archive
Measured across GPT-5.6, Claude Fable 5, Qwen3.7-max and five more families: reasoning dominates the output bill. How to read every usage field, and cap them.
Vendors publish a prompt-cache token minimum. Measured across LLM families, auto-cache needs 1.4–2.4x more than the docs say; Claude's explicit cache is exact.
LLM output prices fell about 94% since 2023: how to cut your AI bill without losing quality...
Kimi K3 for coding teams in 2026: benchmarks, real cost and adopt-vs-wait Summary. On July...
A spreadsheet-ready expected-cost model for agent workflows that includes retries, tool-call fanout, context growth, and caching. Plus hard budgets you can actually enforce.
GLM-5.2 self-hosted vs API in 2026: what an open-weight coding agent really costs Summary....
Concrete per-task cost breakdown for Aider, Claude Code, and OpenHands — covering token overhead per PR, monthly burn at team scale, and the break-even formula for hosted APIs vs local models.
GPT-5.5 bills fewest tokens for European languages, Kimi for Chinese, DeepSeek for Japanese; Claude Fable 5, Opus 4.8 and Sonnet 5 run 1.2-2.3x. Measured.
How naive LLM integration drains your SaaS budget and how to fix it with caching, batching, and model routing.
Multi-agent token cost explodes without accounting. The WOWHOW Cost-Attribution Ledger assigns spend per phase, agent, and tool-call with worked numbers.
When LLMs run inside your product, flat-rate pricing stops working. Here's the framework for building AI-native tiers with real inference co
Guy Kobrinsky | Software Engineering Manager @ Meta. Building SemanticGuard At Meta, and in my...