Keep pulling the thread on Computer History.
An AlphaSense study found that switching to cheaper Chinese AI models does not necessarily result in cost savings for real-world financial analysis tasks.
In an AlphaSense study, GPT-5.6 SOL completed financial analysis tasks for 13% less cost and with a 20% higher quality score than QIMI K3.
More expensive models based on price-per-token can be less costly overall because they are more efficient in their token usage for a given task.
OpenAI has introduced an "Ultra Fast Mode" for GPT-5.6 Sol that runs at 750 tokens per second.
OpenAI's "Ultra Fast Mode" leverages optimizations for running models on Cerebras hardware.
OpenAI's Chief Revenue Officer, Denise Dresser, is leaving the company after a 9-month tenure.
Dolly Rajak, the former president and COO of Wizz, has been announced as the new Chief Revenue Officer of OpenAI, replacing Denise Dresser.
Grokbot allows users to teach it a task by manually recording their actions in a browser.
ChatGPT's "Computer History" feature learns how a user works over time by observing their computer usage.
In testing by Artificial Analysis, the Gemini 3.7 Flash model ran at 340 tokens per second.
The Gemini 3.7 Flash model is more than twice as fast as GPT-5.6 Luna and slightly faster than NVIDIA's NeMoTron 3.5 Lightning.
Google's Gemini 3.7 Flash model improved its score on the DeepSWE coding benchmark to 65.3% from 48.6% for the 3.6 Flash model.