Stack Log — 11 Feb 2026
Quick tooling update. Rate limits shifting underfoot, models finding their niches.
Kimi K2.5: Killer Deal Right Now
Burned through the monthly quota in a few days. Expected to be locked out until March. Instead, a week later we’re back in action with weekly limits showing—29% usage, resets in 125 hours. Possible the monthly display was a UI bug, or they restructured tiers silently. Either way, the work continues.
Kimi K2.5 remains the best bang-for-buck in the stack. Frontier capability at prices that don’t punish experimentation.
See /value/kimi-access/ for current access paths.
GPT 5.3 Codex: Solid, Not Flashy
Codex with medium-thinking has been solid. Fast enough for iterative work, capable enough for most architecture decisions. GPT 5.3 gets the job done—it’s not stealing the spotlight, but it’s not causing headaches either. Reliable second chair.
Compare: /compare/codex-vs-claude-vs-kimi/
Opus 4.6: The Deadline Tax
Opus 4.6 still drains tokens fast, but the speed premium is clever product design. When you’re up against a hard deadline—demo tomorrow, client expecting delivery, production incident—those extra tokens for instant responses become trivial against the cost of missing the cutoff. Anthropic knows exactly which mode you’re in when you click that button.
The extended context is genuinely useful for large refactors, but the speed option is the killer feature for deadline mode. You pay the premium when the alternative is failure.
Announcement: Claude Opus 4.6
Pricing breakdown: /compare/claude-vs-openai/pricing/