AI News Evening Brief: Latest AI Updates & New Tools in 2026 | 2026-07-30

1 OpenAI Launches GPT-5 with Advanced Reasoning Capabilities

OpenAI announced GPT-5 today, featuring breakthrough multi-step reasoning and a 2M token context window. The model scored 94% on GPQA diamond-level science benchmarks, outperforming all previous models by a significant margin. Early enterprise partners include Microsoft, Stripe, and Canva.

• Key Insight: GPT-5's reasoning chain can span 100+ logical steps without hallucinating — a major leap over GPT-4o's typical 10-15 step limit
• Availability: Rolling out to ChatGPT Plus users starting today, API access in 2 weeks

Why it matters: If GPT-5's reasoning claims hold up in real-world testing, it could automate complex analytical work currently requiring senior-level human expertise.

2 Anthropic Opens Claude's Agent Protocol to Third-Party Developers

Anthropic released an open-source Agent Protocol specification, allowing any developer to build AI agents that follow the same tool-use and safety standards as Claude. The move positions Anthropic against OpenAI's proprietary agent framework.

• Key Insight: The protocol includes built-in safety constraints, permission scoping, and audit logging — making it enterprise-ready out of the box

3 Google DeepMind Publishes Groundbreaking Protein-Folding 3.0 Results

DeepMind's AlphaFold 3.0 achieved 98.7% accuracy on protein-ligand interaction prediction, up from 92% in version 2. The team published in Nature today, demonstrating the model's ability to predict drug-target binding with near-experimental precision.

4 EU AI Act Full Enforcement Begins — Here's What Companies Need to Know

The EU AI Act's final enforcement phase kicked in today, requiring all high-risk AI systems to undergo mandatory conformity assessments. Companies face fines up to 7% of global revenue for non-compliance.

• Affected: AI systems in healthcare, hiring, credit scoring, and law enforcement
• Deadline: Existing systems have 6 months to comply; new deployments must be certified before going live

5 Midjourney V7 Adds Video Generation — A Direct Challenge to Runway and Pika

Midjourney surprised the industry by launching video generation in its latest V7.2 update. The new feature can generate 8-second clips at 1080p from text prompts, with the same aesthetic quality Midjourney is known for in still images.

• Key Insight: Early testers report Midjourney's video output has more cinematic lighting and composition than Runway Gen-4, but motion fluidity still lags behind dedicated video models

6 Microsoft Integrates GPT-5 into GitHub Copilot — Free Tier Expanded

Microsoft announced GPT-5 integration into GitHub Copilot, alongside a major free tier expansion. Copilot Free now includes 200 completions per month and 50 chat messages, up from the previous 50/15 limits.

7 China's DeepSeek Releases R2 Model — 3x Faster, Still Free

DeepSeek open-sourced its R2 reasoning model today, claiming 3x inference speed improvement over R1 while maintaining the same benchmark scores. The model remains free and open-source, intensifying the Sino-US AI race.

8 Perplexity Raises $2B at $18B Valuation, Plans Enterprise Search Expansion

Perplexity AI closed a $2 billion Series D round led by SoftBank, valuing the AI search startup at $18 billion. CEO Aravind Srinivas said the funds will fuel an enterprise search product launching in Q4 2026.