July 2026 · By ASR AI Studio
The AI Shift This Week!
The week new models got cheaper, and AI agents learned to work while you're away
THIS WEEK AT A GLANCE
| No. | Category | Story |
|---|---|---|
| 01 | Foundation Models | OpenAI releases GPT 5.6 and launches ChatGPT Work, a new tool for everyday office tasks |
| 02 | Developer AI | SpaceXAI launches Grok 4.5, a coding focused model priced well below Claude Opus 4.8 |
| 03 | Agentic AI | Anthropic brings Claude Cowork to web and mobile so tasks keep running when you step away |
| 04 | AI Video | Alibaba's Wan team releases Wan Streamer v2.2, improving real time AI video without added delay |
01 · Foundation Models
OpenAI Launches GPT 5.6 and Introduces ChatGPT Work for Daily Office Tasks
OpenAI · GPT 5.6 · ChatGPT Work · July 2026
OpenAI released its GPT 5.6 model family to the public on July 9, 2026, after a two week limited rollout to about 20
approved organizations at the request of the US government. The lineup includes three versions: Sol, built for coding
and science work; Terra, aimed at matching the previous top model at roughly half the cost; and Luna, a faster and
cheaper option for everyday use. CEO Sam Altman said Sol handles coding tasks 54 percent more efficiently than
earlier versions, and the company calls it its strongest model yet for cybersecurity work. On the same day, OpenAI also
introduced ChatGPT Work, a companion tool for desktop, web, and mobile built to handle everyday office tasks such
as email and scheduling. ChatGPT Work connects with tools like Slack and Salesforce so it can act directly inside the
apps teams already use.
Our Takeaway: Coding teams and engineering leads get a faster, cheaper option for heavy technical work, while
business teams gain a separate assistant built for daily admin rather than open ended chat. Together the two launches
let OpenAI serve both ends of a company, from writing code to filing routine paperwork, inside one subscription.
02 · Developer AI
SpaceXAI Releases Grok 4.5, a Coding Model Priced Below Rivals
SpaceXAI · Grok 4.5 · July 2026
SpaceXAI launched Grok 4.5 on July 8, 2026, positioning it as a model built specifically for coding, agent based tasks,
and general knowledge work. The company trained it using real developer session data gathered through its partnership
with Cursor, running the training across tens of thousands of specialized computer chips from NVIDIA. Grok 4.5 is
priced well below Claude Opus 4.8 and completes coding tasks using roughly a quarter of the output that Opus 4.8
needs for similar work. It also posted a strong score on the Terminal Bench 2.1 coding test. The model is available now
in Grok Build, the Cursor editor, and SpaceXAI's developer console, though availability in the EU was delayed until
mid July.
Our Takeaway: Developers and engineering teams building coding heavy products now have a lower cost option that
claims to match or beat the leading models on real work. For teams watching every dollar of AI spend, that shifts some
of the competition in coding tools toward price rather than raw capability alone.
03 · Agentic AI
Anthropic Expands Claude Cowork to Web and Mobile
Anthropic · Claude Cowork · July 2026
Anthropic expanded Claude Cowork, its tool for delegating multi step work, beyond the desktop app to web browsers
and to iOS and Android phones. People can now start a task at their desk, close the laptop, and let Claude keep
working in the background on Anthropic's servers. Progress updates and approval requests are sent straight to the user's
phone, so a task does not stall just because the original device is offline. The beta rollout began July 7, 2026, starting
with subscribers on the Max plan, and broader access is expected to follow in the coming weeks.
Our Takeaway: Knowledge workers and managers juggling several projects can now step away from their laptop
without pausing a task in progress. That turns Claude Cowork from a desk bound tool into something closer to an
assistant that keeps moving throughout the day.
04 · AI Video
Alibaba's Wan Team Releases Wan Streamer v2.2, Raising Real Time Video Quality
Alibaba Wan Team · Wan Streamer v2.2 · July 2026
Alibaba's Wan research team released Wan Streamer version 2.2, an update to its real time audio and video AI model.
The new version raises video quality from a low resolution output to a sharper 640 by 368 picture while running at 25
frames per second. It handles audio, video, and text together in one system rather than relying on separate tools
stitched together for speech recognition and video rendering. Total response time, including the time it takes to send
data over the network, stays close to 550 milliseconds, near what a live video call feels like
Our Takeaway: Small studios and indie creators building live avatars or virtual presenters get video quality closer to
a real call without the usual lag that breaks the illusion. That makes real time AI video practical for live use, not just
pre rendered clips.
This week's launches point toward the same shift: AI companies are racing to make their models cheaper to run and
useful beyond a single chat window. Coding tools got faster and less expensive, agent based tools started working
across devices, and video generation moved closer to real time. Together the four updates show AI settling into daily
workflows rather than staying a novelty.
ASR AI Studio
ASR AI Studio · Weekly AI Industry Update · For internal and newsletter distribution