Long-Running Coding Agents Need a Harness, Not Just a Bigger Model
Route long-running coding agents by context, modality, tools, and acceptance checks — then use task state, checkpoints, and enforceable limits to keep the loop recoverable.
ChinaAPI Insights
Data-driven notes on Chinese AI models.
Benchmarks, pricing analysis, agent-compatibility reports, and first looks at new Chinese AI models — based on data measured through the ChinaAPI gateway.
All articles
Latest first.
Every article states its method and data source, and gets updated in place when the data changes.
Chinese LLMs vs. GPT, Gemini, and Claude: Route by the Contract, Not the Logo
A practical decision framework for choosing a Chinese LLM or a direct frontier-provider route by product dependency, task constraints, and accepted-output cost—not a generic leaderboard.
Which Video Model Should You Use? Seedance, Kling, Hailuo, and Wan by Task
A practical guide to choosing current ChinaAPI video models for story, movement, reference control, drafts, and edits—without pretending one model wins every job.
Which Image Model Should You Use? A Practical Seedream and Wan Guide
Choose a current ChinaAPI image model by task shape, reference assets, control requirements, and the cost of getting a usable image — not by a generic quality leaderboard.
Choose a Chinese LLM: Current ChinaAPI Routing Guide
Route current ChinaAPI reasoning models by task shape, context window, modality, and usable-answer cost — not a generic quality leaderboard.
Chinese AI API Pricing Landscape — July 2026
What Chinese LLM, image, and video APIs cost in July 2026 — the sub-$0.20 tier, the 1M-context club, per-second vs per-generation video math, and what to pick by workload.