Samwise Tech/AI/Robotics Newsletter
Saturday, July 25, 2026
Anthropic Launches Opus 5, a Cheaper and Less Restrictive Heavyweight Model
Anthropic launched Opus 5 on Friday, adding the newest version of its heavyweight model lineup. Cheaper and less restrictive than the company’s flagship Fable 5, Opus 5 outperforms it on several benchmarks. The model is “much stronger at verifying its work and iterating carefully until it succeeds,” according to Anthropic. Crucially, Opus 5 is not subject to the 30-day data retention policy that covers Fable and Mythos models. Safety classifiers will engage 85 percent less often than with Fable 5, and a new Automatic Fallbacks beta feature routes flagged prompts to a less powerful model rather than returning an error message to API users.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Tech Giants Urge Against Broad Open-Weight AI Restrictions as US Eyes China Response
A coalition of AI companies including Hugging Face, Meta, Microsoft, Mistral, and Nvidia have signed an open letter urging policymakers against imposing premature restrictions on open-weight AI models. The letter arrives as the Trump administration weighs how to respond to alleged Chinese AI intellectual property theft, including accusations that Moonshot AI distilled Anthropic’s Fable model to train its Kimi K3 system. Signatories argue policymakers must distinguish legitimate techniques like distillation from actual IP misappropriation. Replit CEO Amjad Masad stated: “Banning Chinese open models is as good as banning open models in general.” Notably absent: OpenAI, Anthropic, Google DeepMind, and SpaceX — companies with closed-model business interests at stake.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Reid Hoffman and Mark Pincus-Backed Prentis AI Lab in Talks to Raise $100M
Prentis, a computer-use AI lab co-founded by LinkedIn’s Reid Hoffman, Zynga’s Mark Pincus, and CEO Ritankar Das, is in talks to raise $100 million at a $1 billion valuation, according to sources familiar with the discussions. Launched in April 2026, Prentis trains models to automate routine office workflows — navigating documents, handling insurance claims, and processing customs refunds. The startup’s Hive-32B model claims to outperform OpenAI’s GPT-5.4 and Anthropic’s Claude Opus 4.6 on computer-use benchmarks, at roughly 10 times lower cost per task. Prentis has already signed contracts worth up to $50 million with healthcare and manufacturing customers and projects a $75 million annualized run rate by Q3.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Waymo Mulls Breakup with Uber as Autonomous Vehicle Partnership Sours
Waymo is exploring a breakup with Uber, according to the Financial Times, as tensions mount between the two companies. Waymo has already notified Uber that it plans to offer robotaxis on its own app in Austin and Atlanta, where Waymo vehicles currently appear on Uber’s network, starting January 2028. Uber’s contract covering those markets expires in May 2028. The rift mirrors a split in Phoenix earlier this year. Relations have deteriorated steadily: Uber’s CTO publicly called Waymo robotaxi behavior “scary,” and Uber’s CEO criticized Waymo performance in school zones during a May earnings call. The two companies find themselves on opposing sides of multiple robotaxi regulatory battles.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
OpenAI Ships Micro, a $230 AI Keypad Built for ChatGPT Power Users
OpenAI has shipped its first physical product: Micro, a $230 programmable keypad developed in collaboration with specialty keyboard maker Work Louder. Built to pair with ChatGPT and its Codex coding tool, Micro features six customizable “agent” keys for triggering specific ChatGPT sessions, alongside a voice dictation button and Bluetooth or USB connectivity. TechCrunch’s reviewer found the device genuinely useful once programmed but questioned whether the learning curve justifies replacing a standard laptop workflow. Reddit’s response has been largely skeptical — one user called it “a prank” — while an Aftermath review was blunter still. Micro arrives as OpenAI navigates an Apple trade-theft lawsuit and develops additional consumer hardware.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
ChatGPT Desktop App Gains Voice Control to Manage AI Agents and Code
OpenAI updated its ChatGPT desktop application to add voice control, powered by the company’s new ChatGPT-Live model family. Users can now speak multi-step commands to direct AI agents operating within ChatGPT Work or its Codex coding tool — for example, asking the assistant to create a thread, file a pull request, and debug code in a single voice request. On macOS, the Appshots feature lets ChatGPT Voice see what is displayed on screen. The desktop implementation is more capable than the mobile version, supporting complex multi-step workflows and handling clarification prompts mid-task. Remote access also allows iOS users to invoke Codex voice commands from their phone.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Midjourney Acquires Astrology App Co-Star in Unusual Expansion Move
AI image and video generation lab Midjourney has acquired Co-Star, the social astrology app, according to Bloomberg. Terms were not disclosed. Co-Star, which has roughly 4.3 million monthly active users, uses a combination of AI and human writers to generate horoscopes, compatibility assessments, and birth chart analysis. Midjourney, which currently operates exclusively through a Discord server without a standalone app, absorbs Co-Star’s team of approximately two dozen employees — including consumer app developers who could help Midjourney build its first mobile product. The acquisition fits an eclectic pattern of expansion for Midjourney, which is also reportedly building a medical arm and a spa concept.
Sources: TechCrunch Share ↗ ✉︎ Email 💬 Text
Tech Pulse
Top Frontier Models (SWE-Bench Verified): GPT-5.5 (88.7%) | Claude Opus 5 (★ NEW — Jul 24) | Gemini 3.1 Pro (80.6%)
Top Open Source Models: DeepSeek V4-Pro (76.4%) | Kimi K3 (Moonshot AI) | GLM 5.2 (Z.ai)
Top Small Models (15–50B): Qwen 2.5-32B | Phi-4 (14B) | Gemma 3-27B
Top Edge Models (0–15B): Phi-4-mini (3.8B) | Gemma 3-9B | Qwen 2.5-7B
AI Leaders (est.): NVIDIA ($3.4T) | Microsoft ($3.2T) | Alphabet ($2.5T)
Robotics Leaders (est.): Intuitive Surgical ($160B) | ABB Robotics (div.) | Figure AI ($2.6B priv.)
Curated by JD · samwise.agency
Leave a Reply
You must be logged in to post a comment.