ModelsNemotron 3.5 Content Safety: A 4B Guard Model for Text and Images
NVIDIA's 4B guardrail classifies prompts, responses and images, and can enforce your own written policy. How the two modes work and where it fits in a request path.
ReadNotes on models, agents, and what inference costs
Practical writing for developers building with large language models — how they work, how to pick one, and how to keep the bill predictable.
Picking between models — and testing them on your own work.
Models
ModelsNemotron 3.5 Lightning: A Fast Small Model for Agent Grunt Work
NVIDIA's successor to Nemotron 3 Nano keeps 3B active parameters, adds big agentic gains and a permissive licence. Where it beats larger models and where it trails.
Read
ModelsNex-N2.5-Mini: A 35B Apache-Licensed Agent Model on Two GPUs
Nex AGI's smallest N2.5 model is a 35B multimodal agent model under Apache 2.0 that runs on two H100s. What it is, how to read its benchmarks, and when Pro is worth it.
Read
ModelsNex-N2.5-Pro: Where a 397B Agent Model Earns Its Keep
Nex-N2.5-Pro is a 397B Apache-licensed model for computer-use and coding agents. Where its lead over Mini is large, where it is marginal, and what that means for routing.
ReadModelsNorth Mini Code: Cohere's 3B-Active Agentic Coding Model
Cohere's first agentic coding model is a 30B MoE with 3B active, Apache 2.0 weights and a single-H100 footprint. What its scores mean and how to test it in your own harness.
ReadModelso1-pro: Doing the Arithmetic on $600 per Million Output Tokens
o1-pro is o1 with more compute per answer, priced at ten times o1 and served only through the Responses API. A worked cost example and a rule for when it pays.
Read
Modelso3-mini Guide: OpenAI's Text-Only Reasoner and Its Shutdown
o3-mini brought developer features to a cheap reasoning model in January 2025. What it was good for, what it never did, and why 23 October 2026 is the date that matters.
Read
Modelso3-mini-high Explained: A Model Name That Is Really a Setting
o3-mini-high is o3-mini with reasoning effort pinned to high. What that changes in cost and failure modes, how to measure whether high effort helps, and its shutdown date.
Read
Modelso3-pro Guide: Paying Ten Times o3 for Reliability
o3-pro runs o3 with more compute, takes minutes per answer and costs $20/$80 per million tokens. When that trade made sense, how to call it, and what replaces it in December.
ReadModelso4-mini Guide: OpenAI's Cheap Vision Reasoner and Its Successor
o4-mini matched o3-mini on price and added images, tools and fine-tuning. Where it beat o3 on value, where it fell short on facts, and what to move to before 23 October 2026.
ReadModelso4-mini-high: When High Effort Beats Moving Up to o3
o4-mini-high is o4-mini at high reasoning effort. How it compares with o3 on real cost, why effort compounds in agent loops, and a routing rule that outlasts its shutdown.
ReadModelsOpenAI o1 in 2026: Should You Still Be Calling It?
o1 was OpenAI's first full reasoning model in the API. What it costs, how its hidden reasoning tokens bill, and how to decide whether to keep it or migrate off.
ReadShowing 229–240 of 446 articles