ModelsClaude Sonnet 4.6 Guide: The Last Sonnet on the Old Tokenizer
Sonnet 4.6 brought near-Opus coding to $3/$15 with a 1M window. It also keeps controls newer Claude models removed. When it still beats Sonnet 5.5, and when not.
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.
ModelsModelsClaude Sonnet 5 Guide: Legacy Status, Pricing and Reasons to Stay
Sonnet 5 went from flagship Sonnet to legacy in three months. What it costs, what its numbers look like next to Sonnet 5.5, and the cases where it is still the right pin.
Read
ModelsClaude Sonnet 5.5 Guide: Effort Levels, Benchmarks and Migration
Sonnet 5.5 keeps the $2/$10 price of Sonnet 5 and claims up to 30% lower cost per task. How its five effort levels behave, what the benchmarks say, and what breaks.
Read
ModelsCodestral 25.08 Guide: Mistral's Fill-in-the-Middle Specialist
Codestral 25.08 is a completion model, not an agent. What fill-in-the-middle means, what Mistral claims for this release, the batch endpoint, and when Devstral fits better.
Read
ModelsCommand A Guide: Cohere's 111B Dense Model and Its Licence Catch
Command A is a 111B dense model with a 256K window, built for cited RAG answers and runnable on two GPUs. Where it still fits now that Command A+ exists, and the licence trap.
ReadModelsCommand A+ Guide: Cohere Goes Sparse, Apache 2.0 and 48 Languages
Command A+ is a 218B MoE with 25B active, Apache 2.0 weights and vision. What changed from Command A, the hardware it really needs, and where it sits against the field.
Read
ModelsCommand R (08-2024) Guide: Cheap Cited RAG From a 2024 Model
Command R 08-2024 is a 32B model tuned for cited retrieval answers at $0.15 in and $0.60 out. What the refresh changed, why it is pinned, and when to move on.
ReadModelsCommand R+ (08-2024) Guide: A Legacy Model and How to Leave It
Command R+ 08-2024 is a 104B model now on Cohere's legacy price list at $2.50 in and $10 out. What it did well, what replaces it, and how to migrate without regressions.
ReadModelsCommand R7B: Cohere's Small RAG Model and Its Licence Catch
Command R7B is a 7B, 128K-context model built for the answer step of RAG and for tool use. What it does well, where its short output cap bites, and what the licence allows.
ReadModelsCydonia 24B V4.1 Guide: A Fiction Tune on Mistral Small 3.2
Cydonia 24B V4.1 is TheDrummer's uncensored creative-writing fine-tune of Mistral Small 3.2. What the base gives it, how to run it, and where it falls short.
Read
ModelsDeepSeek R1 in 2026: What the Open Reasoning Model Still Teaches
DeepSeek R1 put an MIT-licensed reasoning model next to OpenAI o1 in January 2025. How it was trained, what the 0528 update fixed, and whether to run it today.
ReadModelsDeepSeek V3 (deepseek-chat): The December 2024 Original
The original DeepSeek V3: a 671B MoE with 37B active, released December 2024. Its architecture, its scores, its licence, and why the deepseek-chat name now means two things.
ReadShowing 25–36 of 446 articles