FundamentalsInference vs Training Cost: Which One Is Actually Your Problem
Training is a one-off capital cost you never pay. Inference is a recurring cost you pay per request. How the two differ and why the distinction shapes pricing.
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.
Fundamentals
FundamentalsInstruction Tuning: How a Text Predictor Becomes an Assistant
A pretrained model continues text; it does not answer questions. Instruction tuning is the small, cheap stage that turns one into the other.
Read
GuidesJetBrains LLM Setup: Custom Endpoints in IntelliJ IDEs
How to run a model of your choosing inside IntelliJ, PyCharm or GoLand — what the bundled assistant will and will not do, and the plugin route that does.
Read
GuidesJSON Schema for Tool Calls: What Providers Actually Support
Which JSON Schema features survive the trip to a model, why strict mode changes the rules, and how to write tool parameter schemas that validate and get used correctly.
Read
GuidesKilo Code Setup Guide: Custom Endpoints and Profiles
Configure Kilo Code against your own OpenAI-compatible endpoint, set the model metadata it cannot infer, and assign different models per mode with profiles.
Read
ModelsKimi K2.6 Guide: Specs, Pricing and Where It Still Wins
Moonshot shipped K3 four months after K2.6, but the older model did not become useless. What K2.6 is, what it costs, and the jobs it is still the right pick for.
Read
ModelsKimi K2.6 vs DeepSeek V4 Flash: Four Axes That Decide It
Two models released three days apart with opposite design goals. Active parameters, context length, licensing and vision decide which one fits your workload.
Read
ModelsKimi K2.6 vs DeepSeek V4 Pro: Sight or Depth
K2.6 sees images at 256K context under a custom licence. V4 Pro is text-only, MIT, 1M context, and reasons harder. Two models that barely overlap.
Read
ModelsKimi K2.6 vs GLM-5.2: Vision and 256K, or MIT and 1M
K2.6 sees images and stops at 256K context under a custom licence. GLM-5.2 is text-only, MIT, and takes 1M tokens. Two clean trade-offs, no overlap.
ReadModelsKimi K2.6 vs MiniMax M3: Two Ways to Do Multimodal
Both take images natively, but one gives you 256K of context at a known price and the other 1M at a price sources disagree on. How to pick between them.
Read
ModelsKimi K2.6 vs Qwen 3.6: Hosted MoE or a Single GPU
A trillion-parameter vision model you rent against a dense 27B you can own. The comparison is about deployment shape, not a few points of benchmark difference.
ReadModelsKimi K3 vs DeepSeek V4 Flash: The Widest Gap in Open Weights
The strongest open-weight model against one of the cheapest, both with 1M context. When a twenty-fold price difference is worth paying and when it is waste.
ReadShowing 469–480 of 731 articles