FlintAPI Blog
Chinese AI models, API guides, and developer tutorials.
Two ways model markup reaches your users: leaked DSML and a grammar deadlock
A DeepSeek V4 DSML leak and a GLM-5.3 grammar deadlock look like broken output. Neither is. Minimal repros, three candidate causes, and how to tell them apart.
Import concentration, four cases: Vietnam phones, Indonesia fuel, Thailand chips, Germany cars
Four 2023 import markets from UN Comtrade. Named-supplier concentration runs from 85% down to 10% - and in two of the four markets the figures cannot even name the largest supplier, which is itself worth knowing before you write an email.
How to size an import market with UN Comtrade (worked example: Indonesia, HS 2106)
A step-by-step method to size an import market from the free UN Comtrade API, with a full worked example (Indonesia, HS 2106, 2023) whose every figure you can re-check in one call.
Reading import partner concentration before you pick a market
Market size is the easy number. Partner concentration tells you whether a new entrant can actually get in — worked example with Indonesia versus Vietnam on HS 2106 (2023).
What Chinese LLM APIs actually cost in 2026 (measured per 1M tokens)
A dated price table for twelve Chinese models, why list price is not your bill, and the one line of code that shows your real number.
Why our health check said "up" while the model was gone
Our status page was green for weeks while generations were failing. The check was asking the wrong question.
What actually broke when I put four Chinese LLM providers behind one API
Timeouts, empty 200s, and why a per-attempt deadline beats a bigger one.
The character limit that silently killed our AI replies
We pasted our whole knowledge base into the prompt. The endpoint refused it. Our code answered with a template instead.
Sample deliverable: deciding between a price increase and a cheaper tier
A real, unedited consulting output — the full report our pipeline produced for a pricing decision question on a 30k MRR B2B SaaS, with honest timing and cost notes.
Selling tokens below cost — what a margin audit found in our own gateway
We audited our own rate card against upstream list prices and found five models priced below cost, one of them by about two to one. Here is the arithmetic, the mistake that caused it, and the checks we now run.
What we tell users when 9 of our 13 models are down
A gateway's worst failure is not a crash, it is a model that answers 503 for a week while the dashboard says "up". Here is the contract we now publish, the mistake that caused a silent outage, and the exact 503 body we return when a model is unavailable.
The Chinese LLM Landscape in 2026: A Practical Guide
DeepSeek, Qwen, GLM, Kimi — what each is best at, and how to access them all from one API.
Build a Multimodal App in an Afternoon with FlintAPI
Image, video, audio, and text — one API key, one endpoint style, four modalities.
How Flint-Smart Routing Can Cut Your AI Bill
Automatic model selection routes every prompt to the cheapest model that does the job well — without you thinking about it.
DeepSeek V4 Pro on FlintAPI: Quick Start
Get started with DeepSeek V4 Pro on FlintAPI in under 5 minutes, with a model comparison guide.
Why Your AI Stack Needs a Unified API in 2026
Managing 15 AI provider accounts is a mess. FlintAPI unifies a dozen models behind one API key.
Flint-Smart: The Model That Picks the Best Model
Stop choosing models manually. Flint-Smart analyzes your prompt and routes to the best model automatically.
How FlintAPI's Smart Router Solves Real Problems with Chinese LLMs
Most AI routers just pick a model. Flint-Smart decomposes complex problems, executes subtasks in parallel, and synthesizes answers — all behind one API call.
Drop-in Replacement: Migrating from OpenAI to FlintAPI in 5 Minutes
Swap out one base URL and gain access to a dozen Chinese LLMs. A step-by-step guide to migrating your AI-powered application from OpenAI to FlintAPI.