Research library
Software engineering
Practical architecture and development decisions for software that people can use, maintain, and adapt.
89 articles · Page 1 of 8
Granite Switch 4.1 Combines Chat, Decisions and Speech
IBM combines chat, decision adapters and English transcription in Granite Switch 4.1. What the shared deployment offers, and what to test.
Anthropic's OSS Scanner Sends Free Vulnerability Reports With No Human Review
Anthropic's free OSS Scanner emails unreviewed, model-written bug reports to maintainers. How to enroll, what 88% and 90% mean, and disclosure terms.
Claude Managed Agents Adds Server-Run Workflows and OpenTelemetry Tracing
Anthropic's Managed Agents can now run fan-out workflows on its servers and export traces over OTLP. Limits, billing, content defaults and what to test.
Gemini 4 Argon Costs $2/$10 Until November 30, and Old Gemini Code Needs Changes
Gemini 4 Argon is on the paid Gemini API at $2/$10 per 1M tokens until Nov 30, then $4/$20. Explicit caching is gone and sampling params are ignored.
Gemini API Reroutes 3.7 and 3.5 Flash Calls, and Deep Research Pro Preview Can End October 23
Since October 8, gemini-3.7-flash and 3.5-flash calls are served by 3.8 and 3.6 Flash. What changes, what to retest, and the Deep Research deadline.
GitHub Actions Lockfiles Pin Every Dependency to a Commit. They Don't Vet Any of Them.
GitHub's Actions lockfile preview pins actions, transitive deps and reusable workflows to commits. What it fixes, what it leaves to you, and what to test.
Liquid LFM2.5-24B-A2B: A Laptop-Sized Agent Model With a $10M Revenue Line
Liquid's LFM2.5-24B-A2B runs 2.45B active parameters and leads its own tool-use tables. It is free for commercial use only below $10M annual revenue.
NVIDIA RTX Spark and Nemotron 3.5 Bolt: A 74B Model in 24GB, Before Anyone Can Test It
NVIDIA's Bolt packs a 74B-A9B MoE into 24GB with 2-bit QAT; RTX Spark offers up to 128GB. What fits, what's unpublished, what to test first.
LightOnOCR-3 Adds Layout and Chart Extraction
LightOn's three OCR models add document regions, image descriptions and chart data. How to evaluate the output before using it in retrieval.
Meta Releases Muse Wonder and Muse Wonder Full
Meta's Muse Wonder models target sustained agent work. What is available now, what Active Memory means, and what teams should test.
Anthropic's Cyber Verification Program Now Has Three Tiers. Which One to Apply For
Anthropic split its Cyber Verification Program into Defense, Red Team and Specialized tiers. Who qualifies, review times, the retention catch and Bedrock.
Claude Haiku 5.5: A 90% List-Price Cut, a 75% Real One, and Five Ways to Get a 400
Haiku 5.5 lists at a tenth of Haiku 4.5's price; Anthropic says about 75% less in practice. The gap, the 400s after switching, and Sonnet 5.5 vs Luna.