Daily AI Brief: August 16, 2026
Daily brief: DeepSeek V4-Pro GA with peak/off-peak pricing, Gemini 3.7 Flash at half price, GPT-5.6 Sol Ultrafast, Anthropic risk report, GLM-5.3.
Today
DeepSeek V4-Pro is GA; peak/off-peak pricing begins Saturday. Terminal-Bench 2.1 at 87.9; new rates land August 16 at 16:00 UTC. Why: agent speed meets dynamic pricing. DeepSeek
Gemini 3.7 Flash ships at half the price of 3.6 Flash. GA at $0.75/$3.75 per million tokens through December 31; FrontierCode jumps to 43.6%. Why: the workhorse tier gets cheaper. Google
GPT-5.6 Sol hits 750 tokens/second on Cerebras. Ultrafast runs up to 14x faster than Standard in a limited API preview. Why: frontier speed without the intelligence tradeoff. OpenAI
Anthropic’s August Risk Report keeps risk at “low.” Misalignment risk nudged up from “very low”; a year-long bio-classifier gap disclosed. Why: safety claims get audit trails. Anthropic
GLM-5.3 sets open-weight coding records. Terminal-Bench 3.0 jumps 4.6 to 28.3; weights promised in two weeks after safety checks. Why: post-training beats pretraining scale. Z.ai
Anthropic explains text watermarking. Claude watermarking complies with the EU AI Act with no quality impact or extra tokens. Why: AI content provenance becomes standard. Anthropic
Our take
August 13–16 was speed week: DeepSeek shipped GA, Cerebras ran OpenAI’s flagship at 750 tokens per second, and Google halved workhorse pricing. Related: DeepSeek V4-Pro deep dive, Ultrafast deep dive, GLM-5.3 deep dive.