DeepSeek just put its V4-Flash model into public beta through its API, and the pricing is the part that should get your attention. It’s running at 14 cents per million input tokens with a full million-token context window. That undercuts OpenAI’s newly discounted budget tier by roughly a third, and it’s a fraction of what the smaller Western models charge.
The more interesting part isn’t even the price. DeepSeek says V4-Flash beats its own higher-tier preview model on coding benchmarks like Terminal Bench and DeepSWE, despite sharing the same architecture and just going through another round of post-training. That’s the quiet story in AI labs right now: companies are squeezing more capability out of models they already built instead of always training bigger ones from scratch.
It also plugs straight into the Responses API format and works with Codex out of the box, which is DeepSeek clearly trying to make switching over as easy as changing one line in your code.
Two days before this, OpenAI cut prices on two of its own GPT-5.6 tiers by up to 80 percent. That wasn’t a coincidence of timing so much as a pattern. Every lab is racing to make its models cheap enough that cost stops being the reason you’d pick a competitor. For developers and small teams building on these APIs, that’s a genuinely good problem to have. A year ago, running serious AI features at scale meant budgeting carefully around token costs. Now the fight to be your cheapest option is doing a lot of that budgeting for you.
Competition between labs usually gets framed as a race that ends badly for someone. Here it mostly just means cheaper, faster tools land in your hands sooner than expected. I’ll take that trade every time.
——
Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277
Sources:
https://www.bloomberg.com/news/articles/2026-07-31/deepseek-unveils-public-beta-api-for-flagship-ai-model
https://technode.com/2026/07/31/deepseek-puts-v4-flash-api-into-public-beta/
https://www.odaily.news/en/newsflash/505459
Repost this. Thanks.