Haiku 5.5 costs a tenth of Haiku 4.5 per token, but a new tokenizer and a 100k step change the bill. A worked example and…
ThinkingBox grades AI agents on the database they leave behind. Here is what its numbers show about silent failures, plus a Postgres harness to test…
I ran Prism's ternary-quantized Bonsai 2 27B, a 27B model squeezed under 6GB, on a GPU-less VPS to test its near-lossless claim. Here's what happened.
Opus 5.5, GPT-6 Sol and Luna, and Grok 4.7 all repriced this week. I ran the per-call math with cached tokens included, and the cheapest…
How to build AI guardrails with Jev: one cheap screening call per message, thresholds in code, Next.js and Laravel examples, and the risks the docs…
Jev's 444x cheaper claim, checked against TypeSafe's own evals: which model it's compared to, where Jev loses on invoices, and what the accuracy numbers measure.
A hands-on TypeSafe AI tutorial for Jev: the real request shape, Choice vs Score vs Noul, how confidence is computed, and the failure modes the…
Jev returns probabilities instead of text, at $0.042 per million input tokens. Why that fits my ticket triage bot better than a chat model, and…
A new arXiv benchmark shows the same 7B model scoring 0.83 or 0.27 on receipts depending only on the OCR engine. Why I now test…
A new paper says LLMs catch 60% of JavaScript vulnerabilities and SAST catches zero. The zero is rigged. The per-CWE table underneath is what I'd…