above.dev Blog
Guides on OpenAI-compatible integration, model routing, token spend, long context, and building with above.dev.
Free keys now call DeepSeek V4.1 Flash as well as GLM 5.3 Flash, with vision and a 1M context. Plus what 3.14 billion tokens of real agent traffic says about caching.
Read articleWe gave 133 accounts $10 of free API credit and logged every request. $1,330 issued, $155.80 spent, 96% of input tokens served from cache, and what that says about rate limiting agent traffic.
Read articleIn three days DeepSeek announced the end of V4 Pro, postponed it, then cancelled it "in response to user demand". V4.1 Flash still launched at a lower price. What changed, what did not, and what the reversal says about how people actually pick models.
Read articleAgents re-send their context every turn, grow it all session, run unattended, and spend in a loop you cannot see. Four things that shaped how above.dev is built.
Read articleGLM 5.3 Flash is cheaper than DeepSeek V4 Flash on input and output at any hour, with vision and a 1M context. The $10 giveaway that ran alongside this post is now closed.
Read article