ModelsGPT-6.1 Sol Pro: Pro Reasoning Mode, Explained and Costed
GPT-6.1 Sol Pro is the same model as GPT-6.1 Sol run in pro reasoning mode. How mode differs from effort, why the rate card hides the cost, and how to test it.
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.
ModelsModelsgpt-oss-120b: OpenAI's Open-Weight Model on a Single 80GB GPU
gpt-oss-120b is a 117B MoE with 5.1B active, Apache 2.0 weights and MXFP4 quantisation. How to run it, the harmony format trap, and where it falls short.
Read
Modelsgpt-oss-20b: A Reasoning Model That Runs in 16GB of Memory
gpt-oss-20b has 21B parameters, 3.6B active and Apache 2.0 weights that fit in 16GB. What it does well, where its thin world knowledge bites, and when to step up.
ReadModelsgpt-oss-safeguard-20b: Moderation From a Policy You Write
gpt-oss-safeguard-20b classifies content against a policy supplied at inference time. How to write that policy, pick an output format, and keep the latency bearable.
Read
ModelsGranite 4.0 H-Micro: IBM's 3B Hybrid Mamba Model Explained
Granite 4.0 H-Micro is a 3B Apache 2.0 model with 36 Mamba-2 layers and 4 attention layers. Where the hybrid design saves memory, and where it does not help.
Read
ModelsGranite 4.2 8B Guide: IBM's Dense Reasoning Model, Three Modes
IBM Granite 4.2 8B is an Apache 2.0 dense model with thinking, non-thinking and low-effort modes in one checkpoint. Its benchmarks, its costs and how to run it.
Read
ModelsGrok 4.20 Guide: One Model Name, Three Different Endpoints
Grok 4.20 ships as a reasoning model, a non-reasoning model and a multi-agent variant at one price. What each one is, the 200K pricing cliff, and when to pin it.
ReadModelsGrok 4.20 Multi-Agent: Parallel Research, Billed Per Agent
Grok 4.20 Multi-Agent runs 4 or 16 agents on one request and returns the leader's answer. How it works, what it cannot do, and why the per-token price misleads.
Read
ModelsGrok 4.3 Guide: The Cheap 1M-Context Tier of the Grok Family
Grok 4.3 keeps a 1M window at $1.25/$2.50 per million tokens, adds five effort levels and a 20%-off batch endpoint. Where it fits below Grok 4.5 and when to step up.
ReadModelsGrok 4.5 Guide: xAI's Turn Towards Agentic Coding
Grok 4.5 traded a 1M window for a 500K one and a coding-first training run. Its benchmarks, the token-efficiency claim worth testing, and when it still makes sense.
ReadModelsGrok 4.6 Guide: Long-Running Agents and Dearer Cache Reads
Grok 4.6 improved on Grok 4.5 across its launch table at the same $2/$6 price, but cached input rose to $0.50. What changed, where it trails, and if it still fits.
Read
ModelsGrok 4.7 Guide: xAI's Flagship, and What Changes on Upgrade
Grok 4.7 is xAI's current flagship: a larger base model at Grok 4.6's price. Its benchmarks, where it trails, and the integration details that bite when you switch.
ReadShowing 133–144 of 446 articles