AI News
Daily Tech Reader - AI Edition
Podcast 🎧 • Video 📽 • Short 📽
Claude
Claude
Claude Code Auto Mode Becomes Default August 14 — Pro, Max, and Team Plans Included
Anthropic will make auto mode the default for new Claude Code sessions on Pro, Max, and Team plans starting August 14, citing that developers already approve 97% of permission prompts, and will stop charging extra for the safety classifier compute.
Sources: Anthropic · Help Net Security
Anthropic's Claude Improves Riemann Hypothesis Lower Bound — From 41.6% to 67.2%
An unreleased research version of Claude improved the longstanding lower bound on the fraction of Riemann zeta zeros satisfying the Riemann hypothesis, synthesizing recent papers to push the figure from 41.6% to 67.2%.
Sources: AI Weekly
Claude Outage Reported This Morning — DownDetector Shows Spike Around 6:10 A.M. Eastern
Claude users reported access issues Monday morning, with DownDetector FR logging a rise in outage reports around 12:10 p.m. CET (6:10 a.m. Eastern), as users shared issues on Statusgator.
Sources: DownDetector · DesignTAXI Community
ChatGPT
OpenAI Retires GPT-5.2 and GPT-5.3 API Models Today — Teams Must Migrate to GPT-5.6 Sol
As of August 10, gpt-5.2-chat-latest and gpt-5.3-chat-latest are removed from the OpenAI API; requests still pinned to either model ID will stop working, with GPT-5.6 Sol as the recommended replacement.
Sources: OpenAI · AI Models Navi
ChatGPT Voice Adds File Uploads and Projects — Enterprise and EDU Get GPT-Live as Default
GPT-Live in ChatGPT Voice now supports file uploads and Projects, letting users analyze documents mid-conversation; GPT-Live is now the default voice experience in eligible Enterprise and EDU workspaces.
Sources: OpenAI · Releasebot
ChatGPT Error Rates Spike Monday Afternoon — OpenAI Reports Mitigation Applied by 3 P.M.
OpenAI's status page logged increased error rates for some ChatGPT users Monday, with mitigation applied at 3:07 p.m. and full recovery confirmed by 8 a.m. the same morning for a separate earlier incident.
Sources: OpenAI Status
Gemini
Fortune Publishes Inside Account of How Hassabis Lost the DeepMind CEO Role
A Fortune exclusive today details how low morale, delayed models, and Hassabis's drift away from day-to-day duties contributed to his departure, with Gemini 3.5 Pro still months behind after missing three release deadlines.
Sources: Fortune
Koray Kavukcuoglu Now Runs Google DeepMind — Jeff Dean Departs After 27 Years
Hassabis officially stepped down August 5; Kavukcuoglu takes over Gemini model development as SVP reporting directly to Sundar Pichai, while Jeff Dean leaves to co-found Discovery Loop with Sanjay Ghemawat and Oriol Vinyals, backed by Google.
Sources: Fortune · Axios · Gizmodo
Gemini 3.5 Pro Months Behind Schedule — Engineers Cite Failure to Prioritize Coding Abilities
Three DeepMind engineers told Fortune the flagship model has missed its June target and two subsequent deadlines, blaming the company's failure to prioritize AI coding skills at a time when rivals have made it central.
Sources: Fortune
Copilot
Grok Imagine Video 1.5 Now in xAI API — Text-to-Video and Native 1080p Generally Available
xAI's image reference system, text-to-video, and native 1080p support are now live in the API via grok-imagine-video-1.5, with voice reference support available on request for locking both face and voice across scenes.
Sources: xAI · Releasebot
NVIDIA Ships Alpamayo 2 Super for Commercial Use — Open 34B Autonomous Driving Model
NVIDIA released Alpamayo 2 Super under the permissive OpenMDW-1.1 licence, a 34B open driving model scoring 79.2 on LingoQA — beating Gemini 2.5 Pro by 15 points — with weights available on Hugging Face.
Sources: NVIDIA · AI Models Navi
Grok
Grok Build Adds Custom TLS Certs, Faster Terminal Resize, and Marketplace Owner Wildcards
The latest Grok Build update introduces custom TLS root certificate support, faster terminal resizing, owner-level wildcard entries for marketplace allow/block lists, and a /teleport hint for moving cloud sessions locally.
Sources: xAI · Releasebot
Grok Voice Think Fast 2.0 Alias Goes Live — Priced at $0.08 Per Audio Minute
The grok-voice-latest alias switched to Think Fast 2.0 on August 5, raising the rate from $0.05 to $0.08 per audio minute, while Grok Imagine Video 1.5 adds reference-to-video capped at 720p alongside full 1080p text and image generation.
Sources: xAI · AI Pricing Guru
China AI
Alibaba's Qwen3.8-Max Tops Chinese Leaderboards — Weights Drop for Download This Week
Qwen3.8-Max, Alibaba's 2.4-trillion-parameter open-weight model, ranked highest among Chinese text models on Arena.AI and second globally in vision tasks, with public weight downloads scheduled for this week.
Sources: Alibaba · Outlook Business · CTech
DeepSeek V4-Flash Is Now the World's Cheapest Major Model — $0.03 Per Benchmark Test
Artificial Analysis benchmarks show DeepSeek's V4-Flash averaging $0.03 per test — well below Moonshot's Kimi K3 at $0.86 and GPT-5.6 Sol at $1.86 — while matching OpenAI's budget tier on quality at 40% lower cost per task.
Sources: Artificial Analysis · The Register · Daily Sabah
China's Open-Weight Offensive Accelerates — Moonshot, Alibaba, and DeepSeek All Attack on Price and Scale
With Kimi K3 at 2.8T parameters, Qwen3.8-Max at 2.4T, and V4-Flash optimized for Huawei Ascend chips, Chinese labs are now challenging U.S. frontier models on both benchmark performance and deployment cost simultaneously.
Sources: Cybernews · The Register
Big Story
The DeepMind Reckoning — Missed Models, Talent Exodus, and What Comes Next for Gemini
Fortune's inside account published today reveals the full scope of DeepMind's difficulties: Gemini 3.5 Pro is months overdue, morale is low, and the lab has lost Hassabis, Jeff Dean, John Jumper, and Noam Shazeer in a matter of months — raising urgent questions about Google's AI standing.
Sources: Fortune · Semafor · Axios
Claude Code Auto Mode by Default — Anthropic Bets That Autonomous Agents Are Ready for Primetime
Making auto mode the default for three paid tiers — and removing the safety classifier surcharge — signals that Anthropic believes agentic, low-interruption AI development is now mainstream, not experimental.
Sources: Anthropic · Help Net Security
Business
OpenAI Adds Nubank and BNY CEOs to Its Foundation and PBC Boards Ahead of IPO
OpenAI added David Vélez of Nubank and Robin Vince of BNY to its foundation and public benefit corporation boards, continuing a pattern of institutional credentialing ahead of a Q4 2026 IPO targeting a near-$1 trillion valuation.
Sources: AI Weekly
Anthropic Spent $1.97M on U.S. Lobbying in Q2 2026 — Up 26% Quarter Over Quarter
Anthropic's Q2 lobbying spend outpaced Nvidia and nearly matched Oracle, reflecting intensifying AI policy activity as the company pushes its $40M regulation initiative ahead of the 2026 midterms.
Sources: AI Weekly
Fireworks AI Closes $1.5B Series D — Targets Enterprise Model Specialization
Fireworks AI raised $1.505 billion in Series D funding to help enterprises turn general-purpose AI models into specialized systems trained on their own data, continuing a broader trend of record-scale AI infrastructure investment.
Sources: Crunchbase