In partnership with
Welcome, AI enthusiasts
A trading firm can now run GPT-5.6 Sol fourteen times faster than the developers building on the same model. OpenAI opened that setting on Thursday to a small group of businesses it picked itself, The company says the list will grow as it gets hold of more chips, though it has not put a date on that. Until then the fastest frontier AI belongs to whoever OpenAI decided should have it first. Let's dive in!
In today’s insights:
OpenAI Hands 14x Faster GPT-5.6 to Wall Street
DeepSeek's New Flagship Costs Less Than Claude's Cheapest Model
ChatGPT Now Carries One Memory Between All Your Apps
Read time: 5 minutes
LATEST DEVELOPMENTS
GATED AI ACCESS
🏎️ OpenAI Hands 14x Faster GPT-5.6 to Wall Street
Evolving AI: OpenAI gave Jane Street and a few other firms a version of Sol nobody else can get.
Key Points:
Ultrafast is a new OpenAI service tier that runs the flagship Sol model up to 14 times faster than standard processing.
Jane Street, Rogo, Basis and the voice AI company Podium are among the first businesses running on it.
OpenAI is keeping the preview small for now and says it will add customers as it gets more capacity.
Details:
GPT-5.6 Sol reaches roughly 750 output tokens per second on the new setting, and OpenAI opened the preview yesterday with nothing shipped for ChatGPT alongside it. The chips come from Cerebras, and the setting sits inside the API where companies build their own products on top of Sol. Fast answers have always come from smaller models, so any business putting one into a live product had to accept weaker reasoning to get them. OpenAI now runs Sol on the faster setting during its own outages, where engineers read logs and check traces while the system is still down.
Why It Matters:
Jane Street uses it for the assistants its developers work with, and Podium uses it on live customer calls where a wait of a few seconds costs the business money. The pause that comes before every AI answer is still there for everyone outside that list, in a support chat or on a checkout page. Most people will get this speed secondhand, whenever a company they already deal with decides to pay for it.
TOGETHER WITH DATADOG
📈 How to Drive AI ROI Guide
Evolving AI: Connecting cost, performance and infrastructure so you can scale responsibly.
AI spend is growing fast and for most organizations, it’s growing in the dark. This practical guide is for engineering, finance and FinOps leaders who need more than dashboards in the age of AI; they need answers.
Inside, you’ll learn how to:
Break down AI costs by token, model, provider and team across OpenAI, Anthropic, and beyond
Get alerted the moment inference volume spikes or API spend exceeds budget, not weeks later
Correlate cost increases directly to architectural changes, so root-cause analysis takes minutes, not months
Put cost data in front of the engineers making the decisions that drive it
Connect cloud spend to GPU efficiency and model performance for a complete picture of ROI
See how Kevel cut AWS costs by up to $100,000 per month after replacing reactive cost reviews with real-time visibility enabled by Datadog.
Source: The Economist
Evolving AI: Deepseek V4-Pro sits at the top of DeepSeek's lineup and the bottom of the price board.
Key Points:
DeepSeek has ended a four-month preview and put V4-Pro on its app, web and API.
V4-Pro is built to run agent work unattended, with three reasoning effort levels and one-click setup inside OpenAI's Codex.
Peak and off-peak billing takes effect on August 16, with off-peak set at half the peak price.
Details:
DeepSeek set V4-Pro output at $3.96 per million tokens during peak hours and $1.98 outside them, up from a flat $0.87. Claude Haiku 4.5, the cheapest model Anthropic currently sells, charges $5 per million output tokens. V4-Pro is the largest model DeepSeek has built, with 1.6 trillion parameters and a million-token context window. Reuters put the increase across the V4 line at anywhere between 50% and 1,100%, depending on the token type and the hour of the call.
Why It Matters:
Claude Opus 5 and GPT-5.6 Sol charge $25 and $30 for that same million output tokens at any hour of the day. Agent work burns output by the million, so a long run costs different amounts depending on when it starts. Peak covers 01:00 to 04:00 and 06:00 to 10:00 UTC, so a European morning sits inside the costly window and an American one does not.
Evolving AI: A new Mac setting lets ChatGPT read the work you do outside the chat window.
Key Points:
Computer History logs clicks, typing, keyboard shortcuts and app switches through macOS accessibility.
The feature captures no screenshots, no microphone input and nothing opened in private browsing.
Each person turns it on individually, and Business and Enterprise workspaces need admin approval first.
Details:
OpenAI's Computer History gathers activity from every app and website a person allows, then builds a timeline of day-by-day summaries that ChatGPT and Codex can question later. Raw event files are deleted from the machine after 48 hours, while the summaries remain as plain-text Markdown files that are not encrypted. OpenAI warns that pulling website content into that record raises the risk of prompt injection reaching Codex.
Why It Matters:
Slack conversations and browser tabs now feed the same record ChatGPT reads when your question comes in, so the context arrives without you pasting anything. The assistant only becomes useful this way by watching how you move through your own working day, and every app you allow widens what it sees.
👀 Click on the image you think is real
QUICK HITS
🤝 IBM struck a strategic partnership with OpenAI, embedding GPT-5.6, Codex, and ChatGPT Work into its consulting platform.
🚀 Cognition is in early talks to raise at a $40B valuation, up 54% in three months as Devin revenue nears $1B.
🏗️ OpenAI-backed Thrive Holdings raised $2B at $12B to buy accounting and IT firms and rewire them with AI.
🎭 Spotify will badge "AI Persona" artists in mid-September and cut them from all recommendations by default.
📈 Trending AI Tools
🐰 CodeRabbit - Ship higher quality code with AI-powered code reviews*.
🌐 Nitro - Fast human plus AI translation in 70+ languages, delivered in hours.
🎨 Omniwork - Creative Agent OS where expert AI agents run your whole pipeline.
🛒 Kopai - Turn your expertise into an AI agent you can publish and sell.
*partner link