News & guides
Qwen 3.5 for Coding: Picking a Size You Can Actually Run
The Qwen line is the only open family that spans 27B to 397B under Apache 2.0. A guide to choosing a size, reading its benchmarks, and self-hosting economics.
· 10 min read
RAG vs Long Context: Retrieval Did Not Become Obsolete
Large context windows were supposed to kill retrieval. They did not. Here is how the two actually compare on cost, accuracy and latency — and when to use each.
· 8 min read
Rate Limits and Retries: Backoff That Does Not Make It Worse
Token buckets, jitter, retry budgets and circuit breakers for LLM APIs — how to stay under the limit instead of discovering it, and why naive retries amplify outages.
· 9 min read
ReAct vs Plan-and-Execute: Two Ways to Run an Agent Loop
ReAct decides one step at a time. Plan-and-execute commits to a plan first. The choice changes cost, latency, recoverability and how failures look.
· 9 min read
Reasoning Models Explained: When Thinking Longer Helps
Reasoning models spend extra tokens working before they answer. Here is what that changes mechanically, which tasks it helps, and where it is just an expensive delay.
· 9 min read
Roo Code Setup: Custom Endpoint and Profiles Per Mode
Configure Roo Code against an OpenAI-compatible base URL, then use API configuration profiles to give each mode its own model. Includes the tool-calling caveat.
· 8 min read
Running LLMs in CI Without Flaky, Expensive Builds
Putting a model call inside CI adds a non-deterministic network dependency to your build. How to keep it cheap, secret-safe on forks, and incapable of blocking a merge.
· 9 min read
Self-Hosting vs a Managed API: Find Your Break-Even
Running open weights on rented GPUs has a break-even volume. Here is how to compute yours from throughput and utilisation, and why most teams never reach it.
· 9 min read
Setting an AI Budget for a Small Engineering Team
A bottom-up method for budgeting model spend on a team of five to twenty, including the buffer to hold, the caps to set, and the alerts that matter.
· 8 min read
Small vs Large Models: Why Parameter Count Stopped Meaning Much
Mixture-of-experts split model size into total and active parameters, and only one of them predicts your bill. How to think about size when picking a model.
· 10 min read
Speculative Decoding: Faster Output, Identical Tokens
Speculative decoding drafts several tokens cheaply and verifies them in one pass, cutting latency without changing the output distribution. How it works.
· 8 min readStop Reading Benchmarks. Build a 20-Task Eval Instead.
Public leaderboards tell you almost nothing about your codebase. Here is how to build a private eval set in an afternoon and actually know which model to use.
· 9 min readReady to try it?
Every model behind one key. Start free and upgrade when it sticks.