Making AI Headlines · Page 3
Deepseek Offers an Agent That Gets to the Point
· New York time
Deepseek has released V4.1-Flash-1017, an open-weight model reported to improve benchmarks, finish tasks faster and produce shorter outputs than its predecessor. It is also opening its internal search engine through Search and Fetch APIs, priced at $0.75 per 1,000 calls, with free access for now for API users in tools such as Claude Code and Codex. The combination targets a practical concern in agent systems: a task is not cheap merely because one individual response is cheap. Repeated searches, retries and long explanations can turn a modest request into a small administrative civilization. Shorter outputs appeal to anyone who has asked for a decision and received the literary equivalent of a warehouse. Brevity is useful when it removes ceremony, though less useful when it removes the reason an answer should be trusted. Benchmark improvements provide a starting point for testing real workflows. Teams still need to see whether the model completes their tasks accurately, uses tools sensibly and stops when the job is done. The broader contest concerns the cost of an outcome rather than the grandeur of a demonstration. A capable agent should resemble a good courier: handle the assignment and leave, without delivering a commemorative speech about the journey to the front desk.
Read this story in the full newspaper →Satire. Not factual reporting. Read the original reporting from the linked publishers.