AI BriefingThe AI Middle Tier Is Disappearing
Key point
New pricing policies from OpenAI and DeepSeek have polarized the AI cost structure.
Details
OpenAI released GPT-5.5 at $5/1M tokens input and $30/1M tokens output, doubling the price compared to GPT-5.4. In return, it touts a 1M context and higher task efficiency.
DeepSeek unveiled V4-Pro and V4-Flash in the same period. V4-Pro is priced at $1.74 input and $3.48 output, while V4-Flash is lower at $0.14 input and $0.28 output. Both models are open-weight under the MIT license with a default context of 1M.
The key change isn't the price tags but the market structure.
- The gap between premium closed models and ultra-low-cost open weights has widened significantly.
- Coding agents and inference pipelines will need to redesign their routing strategies per task.
- Workloads are increasingly likely to split between integrated offerings like OpenAI/Anthropic for some tasks and open infrastructure like DeepSeek for others.
According to the article, V4-Pro recorded SWE-bench 80.6%, and support for Huawei Ascend supernodes was also mentioned. Multimodal is not yet supported, so it's expected to be applied first to text-centric workloads.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.