OpenAI Shuts Down Atlas Browser

In partnership with

Speak naturally. Send without fixing.

Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Not rough transcription you have to clean up. Actual polished text — ready for email, Slack, or any app.

Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.

89% of messages sent with zero edits. Millions of professionals use Flow daily, including teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.

Start flowing free

In today's issue

01 · Business

OpenAI shuts down Atlas browser, folds features into ChatGPT and Chrome

OpenAI is shutting down Atlas, its nine-month-old AI browser, and moving its features into a Chrome extension and upgraded ChatGPT desktop app instead. The shift signals that OpenAI now believes ChatGPT's existing user base is more valuable than building a standalone browser.

The details:

The new ChatGPT Chrome extension competes directly with Google's Gemini Side Panel, answering questions about web pages and handling multi-step tasks without leaving the browser.
Atlas lasted nine months after launching in October 2025; the shutdown follows CEO Fidji Simo's directive to cut non-core projects, which also ended the Sora video tool.
OpenAI is building a cloud browser that runs on its servers, letting ChatGPT agents complete tasks on users' behalf while the desktop app stays responsive.

Why it matters: OpenAI is betting that improving ChatGPT where people already use it—inside Chrome and on the desktop—beats the cost of maintaining a separate browser. This avoids the engineering burden of rendering engines and web standards, but leaves Google controlling the extension ecosystem and Perplexity or The Browser Company room to win if they scale their standalone browsers.

Read the full story →

02 · Models

Meta launches Muse Spark 1.1 to challenge Claude and GPT-5.6 on coding

Meta released Muse Spark 1.1 on Thursday, a coding model designed to build features and manage workflows inside enterprise software without human intervention. At $1.25 per million input tokens, it's priced to compete directly with Claude Haiku 4.5 and GPT-5.6 Luna — the dominant cheap coding models that companies are already standardizing on.

The details:

Mark Zuckerberg broke a three-year silence on X to promote the launch, his first post on the platform since July 2023.
Meta also released Muse Image, an AI image-generation model, on Tuesday of the same week — part of a broader sprint to ship new models.
Muse Spark 1.1's real test is quality-per-dollar on actual coding tasks; enterprises already running evals against Claude and GPT-5.6 will have benchmark results within days.

Why it matters: Meta is entering a category Anthropic and OpenAI defined over a year ago, but the pricing is close enough that large-volume enterprise buyers have reason to test it. The model's strengths at tool use and computer use — automating chains of decisions across external apps — are concrete claims that will either hold up or fall short when real teams run it against their existing workloads.

Read the full story →

03 · Business

Sunrun to pay customers to host AI compute nodes in their homes

Sunrun is paying customers to host AI compute hardware in their homes, spreading AI workloads across 1.1 million residential solar and battery systems instead of building massive data centers. This sidesteps a major problem: 70% of Americans oppose new data center construction near them due to noise, pollution, and power drain.

The details:

Sunrun completed a proof-of-concept pilot and plans to resell the aggregated compute capacity to AI companies like Anthropic and OpenAI within the coming months.
Home-based nodes would run partly on rooftop solar and home batteries, reducing grid draw compared to a single warehouse-scale facility drawing hundreds of megawatts.
Major AI labs are racing to secure data center land as local opposition mounts—Anthropic just signed a 20-year Kentucky facility lease, but permitting remains a bottleneck.

Why it matters: Sunrun sees a new revenue stream from customers' existing batteries, while AI companies get around siting backlash—but it only works if home internet is reliable enough and idle nodes don't tank the economics. The real test is whether enterprise buyers will trust distributed residential hardware for anything beyond small-model inference.

Read the full story →

AI Box

Every AI model. One chat.

The latest models from ChatGPT, Claude, Gemini, Sora, ElevenLabs — 80+ models in a single chat. Compare answers side by side. Pick the best one every time.

● ChatGPT, Claude, Gemini, Grok, DeepSeek — in one chat

● Generate images & video with Sora, Veo, Ideogram

● Compare any two models side by side

● From $8.99/mo · 80+ models, all included

Try AI Box →

aibox.ai

Trusted by 3,000+ teams

04 · Business

SK Hynix raises $26.5B in largest-ever foreign IPO on US markets

SK Hynix raised $26.5 billion selling shares on US markets Friday — the largest-ever US debut by a non-American company, beating Alibaba's 2014 record. Wall Street bid aggressively for exposure to the South Korean chipmaker's high-bandwidth memory, which powers Nvidia's AI accelerators.

The details:

The company sold 177.9 million shares at $149 each, priced 2.7% above its Seoul average, with demand more than seven times the available supply.
Stock opened 14% above the IPO price Friday under ticker SKHYV, then began regular trading Monday as SKHY.
US Commerce Secretary Howard Lutnick said he's already negotiating with SK Hynix and Samsung to build new chip factories in the United States.

Why it matters: SK Hynix's record haul signals US investor appetite for AI infrastructure outside America — but also a policy shift. Washington now wants to rebuild domestic memory manufacturing, which it largely abandoned decades ago. Micron fired back with a $250 billion US investment pledge, setting up a real competition for federal subsidies and talent.

Read the full story →

05 · Business

Fidji Simo steps down as OpenAI's applications chief, citing chronic illness

Fidji Simo is stepping down as OpenAI's No. 2 executive and moving to a part-time advisor role after a three-month medical leave. She had consolidated product and business operations under her since May 2025, but a severe relapse of a chronic illness led her to prioritize recovery ahead of OpenAI's planned 2027 IPO.

The details:

Simo was diagnosed with POTS (Postural Orthostatic Tachycardia Syndrome) in 2019 and took medical leave in April 2026 when the condition worsened significantly.
OpenAI has confidentially filed to sell shares on the stock market, targeting a $1 trillion valuation in 2027 — up from its current $852 billion valuation.
Greg Brockman, OpenAI's president, absorbed Simo's product responsibilities during her leave and is now merging ChatGPT and Codex into a single AI agent platform.

Why it matters: Simo's departure removes Altman's designated second-in-command during a critical period of reorganization and IPO preparation. The company will likely promote from within — Denise Dresser, the chief revenue officer hired in December 2025, is considered an internal candidate and brings enterprise sales experience OpenAI needs to compete with Anthropic on coding tools.

Read the full story →