AI News
Daily Tech Reader - AI Edition
Podcast 🎧 • Video 📽 • Short 📽
Claude
Anthropic Sonnet 5 Introductory Pricing Ends August 31 — Standard $3/$15 Rate Takes Effect September 1
Anthropic confirmed the $2/$10 per million token launch price for Claude Sonnet 5 expires August 31, 2026, with standard $3/$15 pricing resuming September 1. Because Sonnet 5's updated tokenizer generates up to 35% more tokens for identical content, effective costs can rise beyond the list-price change alone suggests.
Sources: Anthropic · ClaudeLog · BenchLM
Claude Opus 4.1 Retired August 5 — Developers Directed to Migrate to Opus 5
Anthropic's June 5 deprecation notice takes effect this week, retiring Claude Opus 4.1 on August 5, 2026. Developers still running Opus 4.1 in production are directed to migrate to claude-opus-5.
Sources: Anthropic · Releasebot
Future of Life Institute AI Safety Index: Anthropic Leads at C+, Three Labs Receive Failing Grades
The FLI Summer 2026 Safety Index graded nine AI companies across 37 indicators — Anthropic topped the rankings at C+, with OpenAI and Google DeepMind at C and Meta at D+, while xAI, DeepSeek, and Mistral each received failing marks. Reviewers flagged the industry's pivot toward military AI and a broad retreat from earlier safety commitments.
Sources: Future of Life Institute · AI Weekly
ChatGPT
OpenAI Adds SynthID Audio Watermarking to GPT-Live — Provenance Detection Now Available via API
OpenAI updated GPT-Live so that audio generated through ChatGPT Voice and the API now carries DeepMind's SynthID watermark, with a public verification tool and developer API access for provenance checks launched simultaneously.
Sources: OpenAI
OpenAI Engineering Post: GPT-Live Continuous Voice Architecture — How Full-Duplex Delegation to GPT-5.5 Works
OpenAI published an Aug 3 engineering deep-dive on GPT-Live's full-duplex architecture, detailing how the voice layer delegates to GPT-5.5 for search and reasoning mid-conversation while maintaining conversational continuity, and explaining plans to swap in newer frontier models behind the voice interface as they ship.
Sources: OpenAI Engineering
OpenAI Posts "Building Abundant Intelligence" — Framing Cheap AI Access as the Next Industrial Shift
OpenAI's July 31 company statement positions ultra-low-cost inference — enabled by GPT-5.6 Sol's GPU kernel rewrites — as analogous to the electrification of industry, arguing that making frontier intelligence affordable is the company's primary mission for the next decade.
Sources: OpenAI · Build Fast With AI
Gemini
Google DeepMind Backs $10M Multi-Agent AI Safety Research Call — Deadline August 8
DeepMind, Schmidt Sciences, and partners opened a funding call of up to $10M for researchers studying safety risks in large-scale multi-agent AI systems where models built by different organizations interact across shared digital environments. Proposals are due August 8, with awards expected in fall 2026.
Sources: Google DeepMind · EdTech Innovation Hub
DeepMind Biosecurity Initiative Expands — Isomorphic Labs Partnership Targets Pathogen Surveillance and Vaccine Design
Google DeepMind and Isomorphic Labs launched a bioresilience program supporting pathogen surveillance, vaccine and therapeutic design, and outbreak response, with more than 15 government and research partnerships established. CEO Demis Hassabis simultaneously called for an industry-funded international standards body for frontier AI.
Sources: Axios · IBTimes
Copilot
Microsoft Quietly Tests MAI-Realtime — A Full-Duplex Bidirectional Voice Model Supporting 17 Languages
A hidden entry in Microsoft's MAI Playground reveals MAI-Realtime, a first-party full-duplex voice model with two voices (Victoria and Grant) and support for 17 languages, competing directly with OpenAI's GPT-Live. No public launch date has been announced; limited partner access is already underway.
Sources: TestingCatalog · CryptoBriefing
Microsoft MSFT Gains ~5% Monday — Azure Momentum Extends Post-Earnings Rally Into New Week
Microsoft shares rose nearly 5% in Monday trading, extending last week's record single-day surge as investors continued to price in Azure's crossing of $100B in annual revenue and accelerating AI infrastructure demand.
Sources: Yahoo Finance · AlphaSpace
Grok
xAI Releases Grok Voice Think Fast 2.0 — New Default Voice Model Routes to API August 5
xAI released Grok Voice Think Fast 2.0, its most capable speech-to-speech voice model, with improved intelligence, transcription accuracy, and tool-use reliability. The grok-voice-latest API alias migrates to the new model on August 5; developers staying on 1.0 must pin the previous model string before then.
Sources: xAI · Releasebot
Elon Musk Confirms Grok 4.6 in Two Weeks, Grok 4.7 Two Weeks After That — Both Before End of August
Musk posted to X that Grok 4.6 is approximately two weeks out and Grok 4.7 follows two weeks after, putting both releases before the end of August and maintaining xAI's near-monthly Grok 4-series cadence. Grok 4.5, released July 16, scores 29.0% on SWE Marathon ahead of Claude Opus 4.8 and runs on a 1.5-trillion-parameter V9 foundation.
Sources: Basenor · xAI
China AI
Moonshot Kimi K3 Matches Frontier U.S. Models — Chinese AI Now Dominates OpenRouter's Top 10
Kimi K3, launched July 16, approaches the performance of OpenAI GPT-5.6 Sol and Anthropic Claude Fable 5 on coding and reasoning benchmarks, echoing the DeepSeek moment but at narrower cost margin. At one point in mid-July, six of OpenRouter's top 10 most-used models came from Chinese companies including Moonshot, Tencent, DeepSeek, and MiniMax.
Sources: CNN Business · Fortune · CNBC
DeepSeek V4 Flash Releases in Public Beta — V4 Pro GA and Harness Coding Tool Still Pending Mid-August
DeepSeek released the DeepSeek-V4-Flash API in public beta on July 31, ahead of the originally reported mid-August window. V4 Pro general availability and the in-house Harness coding tool remain pending, with reports placing V4 Pro GA between August 10–20; the old deepseek-chat and deepseek-reasoner aliases were retired July 24.
Sources: DeepSeek · DeepSeekV4Pro.com
Cloud AI
SpaceX Reports First-Ever Public Earnings Tuesday — Starlink, AI Revenue, and Lock-Up Expiration All in Focus
SpaceX posts its first earnings as a public company after Tuesday's bell, with Wall Street expecting $6.9B in Q2 revenue as AI segment sales are projected to jump from $818M in Q1 to roughly $2.2B. The report comes two days before a major insider lock-up expiration, and the stock has already fallen more than 50% from its post-IPO peak of $225.
Sources: Forbes · TheStreet · Yahoo Finance · Morningstar
AMD Q2 Earnings Tuesday — Data Center Revenue Expected to Top 100% Year-Over-Year to $6.5B
AMD reports after Tuesday's bell with analysts expecting $11.3B in revenue and data center growth above 100% to roughly $6.5B, driven by MI350 AI chip shipments. The stock has rebounded toward $500 on Azure and Palantir results but sits about 12% below June highs heading into the print.
Sources: Yahoo Finance · Benzinga · TradingKey
Google Cloud Surges 82% in Q2 to $24.8B — Alphabet Raises Full-Year Capex Guidance to Up to $205B
Alphabet reported Q2 Google Cloud revenue of $24.8B, up 82% year-over-year and its strongest growth in recent quarters, with cloud operating income tripling to $8.8B. The company raised its 2026 capital expenditure forecast to $195B–$205B, citing a supply-constrained AI infrastructure environment and plans to tap third-party cloud capacity as a bridge.
Sources: Yahoo Finance · CNBC · QZ
Big Story
The Voice AI Race Heats Up: OpenAI Has GPT-Live, Microsoft Is Testing MAI-Realtime, and the EU Just Mandated Watermarks on All of It
Three threads converged over the weekend: OpenAI added SynthID watermarking to GPT-Live, Microsoft quietly surfaced a full-duplex voice competitor, and the EU AI Act's audio-provenance rules took effect August 2 — making voice AI the most legally and competitively active frontier in the industry right now.
Sources: OpenAI · TestingCatalog · The Guardian
AI Safety Scores Are In and the Industry Is Struggling — Top Grade Is a C+, Three Labs Are Failing
The Future of Life Institute's Summer 2026 index is a sobering benchmark: nine major labs evaluated, none earned better than a C+, and three — xAI, DeepSeek, and Mistral — failed outright. Reviewers cited military AI expansion, retreating from prior safety commitments, and absent independent audits as the core failures.
Sources: Future of Life Institute · AI Weekly
Business
Palantir Beats Q2 on U.S. Government AI Contracts — Lifts Chip and Cloud Sentiment Monday
Palantir's better-than-expected Q2 results, driven by government AI contract wins, helped restore broader AI market confidence heading into this week's AMD and SpaceX earnings, contributing to Monday's gains across chip and cloud names.
Sources: Yahoo Finance · FX Leaders
AMD Q2 Preview: Can MI350 Shipments and Helios Rack-Scale Orders Push the Stock Above $600?
Analysts at Bank of America and Jefferies raised AMD price targets to $620 and $640 respectively ahead of Tuesday's earnings, arguing AMD has transitioned from GPU supplier to full-stack rack-scale AI system provider. Key focal points: MI350 AI chip revenue, quantifiable Helios orders, and EPYC server CPU margins.
Sources: TradingKey · StockStory
Anthropic's AI for Science Grants Closed Applications Sunday — Up to $50K in Claude Credits for Rare Disease Research
Anthropic's AI for Science grant program, offering up to $50,000 in Claude credits over six months for rare genetic disease researchers, closed applications August 2. Two tracks were available: basic science researchers and early-stage biotechs accelerating clinical development.
Sources: ClaudeLog · Anthropic