News & guides
Flash Attention Explained: Same Maths, Far Less Memory
Flash attention makes long context practical by never writing the score matrix to memory. What it changes, what it does not, and where you feel the difference.
· 9 min read
Forecasting Next Quarter's AI Spend as a Range
Why extrapolating token usage gives the wrong number, which drivers actually move an AI bill, and how to build a defensible range instead of a point estimate.
· 12 min read
Free Tier Strategies: Getting Real Answers Without Paying
Free tiers are evaluation budgets, not production capacity. What they are for, where the limits bite, and how to fit a real model evaluation inside one.
· 12 min read
Function Calling in Python: A Working Implementation
Build a tool-calling loop in Python that survives real inputs: schema generation, argument validation, parallel calls, error returns and the termination condition.
· 9 min read
Function Calling in TypeScript Without Losing Type Safety
A typed tool-calling loop in TypeScript: Zod schemas, a discriminated tool registry, safe argument parsing, streaming assembly and AbortController cancellation.
· 9 min read
GDPval Explained: Benchmarking Real Professional Work
GDPval scores models on deliverables from real occupations, graded by experts. What that measures, why it is not a coding benchmark, and how to read it.
· 9 min read
Generating API Clients With an LLM Without Silent Drift
Where a model beats openapi-generator, where it quietly loses, and how to build a generate-compile-test loop that catches the hallucinated field before you ship it.
· 9 min read
Generating Regular Expressions With an LLM, Carefully
Models write regex fluently and confidently, which is the problem. How to specify the target, demand test cases, avoid catastrophic backtracking and spot dialect mismatches.
· 9 min readGitHub Actions LLM Setup: A Workflow That Reviews PRs
A working GitHub Actions job that calls an OpenAI-compatible model on pull requests — repository secrets, timeouts, rate limits, PR comments and cost control.
· 9 min readGitLab CI LLM Setup: Masked Variables and MR-Only Jobs
A GitLab CI job that calls an OpenAI-compatible model on merge requests — masked variables, rules, artifacts versus MR notes, runner choice and cost control.
· 9 min read
GLM-5.2 vs DeepSeek V4 Flash: Ten Times the Price
Both are MIT-licensed with 1M context, but one costs ten times the other. Where the expensive model earns the gap, and where the cheap one quietly wins.
· 9 min read
GLM-5.2 vs MiniMax M3: Long-Horizon Code or Cheap Multimodal
Both ship 1M context and target coding. GLM-5.2 is built for unattended agent runs; M3 is multimodal and cheaper. Which one your workload actually needs.
· 9 min readReady to try it?
Every model behind one key. Start free and upgrade when it sticks.