Two pricing stories dropped this week that are worth a pause if you're building on LLM APIs.
DeepSeek V4 Flash exits preview — and undercuts its own flagship
DeepSeek V4 Flash left preview at $0.14 / $0.28 per million tokens (input/output) — and it's beating its own larger Pro model on agentic benchmarks, hitting 82.7% on Terminal-Bench. That's a smaller, cheaper model outperforming its own bigger sibling on agent tasks.
Claude Sonnet 5's introductory pricing ends September 1
Anthropic's introductory pricing for Claude Sonnet 5 ends September 1. The headline price rises from $2 to $3 per million tokens — but the bigger change is under the hood: a new tokenizer adds up to 35% more tokens for the same text. Combine the two and the real-world cost increase is closer to 2x, not 1.5x.
What this means if you're building
If your stack leans on frontier models by default, this is a good week to:
- Re-benchmark smaller/cheaper models against your actual task, not just leaderboard scores
- Audit token usage before the September 1 pricing change hits
- Separate "frontier because it's necessary" from "frontier because it's the default"
Cheap isn't always what it looks like on the label, and expensive doesn't always show up as a sticker price change — sometimes it's a tokenizer.
Sources: llm-stats.com, aitoolsrecap.com
Top comments (0)