AI news daily brief: 2026-08-14

Eight stories today: two product moves from OpenAI (a Cerebras-powered inference preview and an ad format expansion), a major model release from DeepSeek, an earnings record from Lenovo, a leadership hire, a Microsoft Copilot release wave, a data center IPO report, and a Cursor update.

OpenAI previews GPT-5.6 Sol Ultrafast mode at 750 tokens per second via Cerebras wafer-scale chips

OpenAI opened a limited API preview on August 13 of GPT-5.6 Sol running on Cerebras wafer-scale chips. The configuration, called Ultrafast, generates up to 750 output tokens per second, 14 times faster than standard GPT-5.6 Sol, per OpenAI. Intelligence is identical to the standard model; only serving speed changes. Target use cases include live voice interactions, customer support, financial analysis, and security incident response. Access is limited to a small group of API customers, with a waitlist open. Pricing has not been disclosed. The arrangement is OpenAI's first public use of a non-Nvidia chip manufacturer for a production-preview API service.

Full story: OpenAI previews GPT-5.6 Sol Ultrafast at 750 tokens per second

DeepSeek V4-Pro exits preview with 87.9 TerminalBench score and open-sources Harness agent framework under MIT

DeepSeek released DeepSeek-V4-Pro-0813 to general availability on August 13 across its app, web, and API. TerminalBench 2.1 performance climbed from 72.1 to 87.9; DeepSWE jumped from 12.8 to 62.7. The model carries a 1 million token context window and supports up to 384,000 output tokens. Alongside the model, DeepSeek published Deepseek Harness under the MIT license, a modular agent framework launchable via npx. A price increase takes effect August 16 at 16:00 UTC, with off-peak rates at half the peak price. At new rates of $0.435 per million input tokens and $0.87 per million output tokens, per DeepSeek, V4-Pro remains roughly 35 times cheaper on input than Claude Opus 4.7.

Full story: DeepSeek V4-Pro exits preview and ships MIT-licensed Harness framework

ChatGPT delivers ads on 1 in 4 commercial queries as OpenAI ships product carousels and oCPC bidding

A study of 50,006 commercial prompts published August 10 by SE Ranking found ChatGPT served paid ads on 25.94% of all commercial queries. Healthcare landed at 28.69%, compared with 2.64% in Google AI Mode for the same category. On August 13, OpenAI added three capabilities to ChatGPT Ads: multi-product shopping carousels drawing from advertiser product feeds, conversion-optimized CPC campaigns in beta, and automatic advanced matching as the default for new tracking pixels. The UK, Mexico, Brazil, Japan, and South Korea joined the geographic rollout on August 11. GEO operators in health and finance should audit organic citation rates now, as paid placements have begun competing with organic results on the same query surfaces.

Full story: ChatGPT serves ads on 1 in 4 commercial queries with carousels and oCPC bidding

Lenovo posts record $26.9B quarter as AI revenue reaches 35% of group sales

Lenovo Q1 FY2026/27 revenue reached $26.9 billion, up 43% year-on-year and the highest quarterly figure in the company's 40-year history, per Lenovo's press release. AI-related revenue through the Infrastructure Solutions Group hit $9.3 billion, or 35% of group sales. Net income grew 61% year-over-year. The result followed Foxconn's August 12 report that AI server revenue crossed 51% of its Q2 revenue. Lenovo is in production ramp for Nvidia Vera Rubin GPU server systems targeted for Q4 2026. The forward signal to watch is Q2 ISG margin: compression below the current 9.1% baseline would shift the story from demand growth to pricing pressure.

Full story: Lenovo posts a record $26.9B quarter as AI server revenue hits 35% of sales

OpenAI appoints Wiz executive Dali Rajic as CRO, second to hold the role in under 12 months

OpenAI on August 13 named Dali Rajic as Chief Revenue Officer, replacing Denise Dresser, who held the post for approximately nine months. Rajic was president and COO at Wiz before Alphabet's $32 billion acquisition of that company, per TechCrunch. OpenAI president Greg Brockman framed the appointment as focused on building "repeatable execution" for the revenue organization. Three major C-suite changes have landed in 90 days: COO Brad Lightcap departed in early August, and AGI Deployment CEO Fidji Simo left in July. Enterprise deal velocity under Rajic and whether executive churn continues into Q4 are the leading signals, given the company's reported IPO preparations.

Full story: OpenAI names Dali Rajic as CRO as C-suite turnover continues

Microsoft 365 Copilot adds governed Agent Store publishing and Anthropic model support in August release wave

Microsoft's August release wave for M365 Copilot introduced admin-gated agent publishing, closing the governance gap that had stalled enterprise extensibility deployments. Custom agents now require IT admin approval before reaching end users. Teams submit agents through Agent Builder; admins review and approve them in the Microsoft 365 Admin Center before they appear under "Built by your org." The approval flow was available from July 15. On August 11, Copilot added Anthropic model support in Word for organizations already running Claude, Outlook coaching improvements, and SharePoint authoritative sites for enterprise search. GitHub Copilot customers managing agents across multiple tenants should review the new Admin Center controls.

Full story: Microsoft 365 Copilot August release: agent governance and Anthropic model support

Vantage Data Centers explores IPO at $100B valuation or sale, sources say

Reuters reported August 13 that Vantage Data Centers is exploring an IPO or sale at a valuation of roughly $100 billion. Vantage is backed by Silver Lake and DigitalBridge. At that figure, a Vantage listing would be the largest data center IPO in history, exceeding Switch's $80 billion target and the $35 billion CoreWeave IPO in March, per Reuters. A listing could raise roughly $10 billion. No banks have been formally retained, and timing and structure remain fluid. The potential transaction arrives as AI infrastructure demand has pushed data center valuations sharply higher across the sector.

Read the full story

Cursor ships agent environments update and launches India Start plan at 649 rupees per month

Per Cursor's changelog, an August 13 update addressed agent environments, noting that agents perform best when a repository is already cloned, dependencies are installed, and setup scripts have run before the agent session begins. Cursor also launched an India-localized Start plan at 649 rupees per month, approximately $7.75, covering access to Cursor models and always-on cloud agents. The August 13 changelog entry was not independently reachable at scrape time; all claims in this section are attributed solely to Cursor's changelog (cursor.com/changelog).