AI Digest of the Week

May 10, 2026 · 6 min read

Perplexity

Top 5 AI Model Releases of the Past Week: Focus on MoE Architectures and Omnimodality

May 3, 2026: Alibaba released Qwen3-Omni — the first open-source omnimodal model, combining text, images, and low-latency multilingual TTS; it leads OpenRouter rankings for reasoning and is available on Fireworks AI.
May 4, 2026: Anthropic introduced Claude Opus 4.7 — an update to their flagship with enhanced agent capabilities, surpassing GPT-5.3 in math and coding benchmarks.
May 5, 2026: Google announced Gemma 4 31B — a powerful MoE model with 31B parameters (4B active), a 256K token context window, and leadership in reasoning tasks, optimized for a single GPU.
May 6, 2026: DeepSeek launched DeepSeek-V4-Pro-Max — a flagship with trillions of parameters, millions of tokens of context, and "fast/expert" modes for complex tasks, confirming insider rumors.
May 7, 2026: OpenAI released ChatGPT Images 2.0 — a top-tier image generation model that instantly topped charts for quality and speed.

Sources:

Habr: Key AI World Events for 1 Week in August 2025
Habr: Key AI World Events for 1 Week in September 2025
Deeplist: Latest News from the World of Neural Networks and AI
AI-Stat: AI Neural Network Release Timeline 2026
Skillfactory: AI News in April 2026
AI-Stat: All AI Models | Complete List of Neural Networks 2025

Gemini

Major AI Releases and Updates from the Past Week: An Overview

This past week has seen several significant developments in the field of artificial intelligence, concerning both new models and the integration of AI into everyday tools.

OpenAI Increases Prices for GPT-5.5 API Access. Starting May 9, 2026, the cost of using the GPT-5.5 model via API has increased by 1.5-2 times compared to the previous version. Despite the price hike, OpenAI claims improved model efficiency, allowing for better results with fewer tokens.
VK Ads Introduces "Creative Studio" with AI Tools. On May 7, 2026, a new section was launched within the VK advertising account, integrating tools for automatic generation and editing of visual materials. The service also includes a "Trend Creative Feed" with examples of effective AI banners.
Apple to Allow Users to Choose Third-Party AI Models in iOS 27. With the release of iOS 27 in the fall of 2026, iPhone users will be able to choose between models from Google, Anthropic, and OpenAI for functions like Siri and text writing tools. Testing of models from Google and Anthropic is already underway.
New AI Model Shows Ability to Detect Early-Stage Pancreatic Cancer. A study published on May 8, 2026, demonstrates that an AI model can detect signs of pancreatic cancer in CT scans up to three years before clinical diagnosis. The model analyzes subtle changes in the organ's structure.

Sources:

Tech.Onliner.by
3DNews
VK Реклама
ZOOM.CNews.ru
AdIndex
vc.ru
Manifold Markets
Sponsr
Samsung Newsroom Россия
ForkLog
Deeplist
DTF
AI на vc.ru
Смартлаб

ChatGPT

Several significant events occurred in the field of artificial intelligence between May 3 and May 9, 2026:

May 1, 2026: OpenAI unveiled GPT-5.5, a new version of its flagship model, with improved text generation capabilities and complex problem-solving abilities.
May 1, 2026: xAI announced Grok 4.3, an updated model focused on deeper contextual understanding and increased response accuracy.
May 3, 2026: DeepSeek released DeepSeek Coder 2.0, an enhanced version of its programming automation model, with improved support for various programming languages.
May 5, 2026: Google Research introduced PaLM-2, a new language model with 200 billion parameters, demonstrating high efficiency in translation and natural language processing tasks.
May 7, 2026: Meta announced LLaMA 3, the third version of its model, with improved code generation and mathematical problem-solving capabilities.

Sources:

TokenBurn.
DeepSeek.
Google Research.
Meta.

Grok

May 5, 2026. OpenAI released GPT-5.5 Instant — a new default base model for ChatGPT that reduces hallucinations in key areas like law and medicine, improves accuracy in math (81.2% on AIME 2025), and adds personalization through conversation memory.
May 6, 2026. xAI introduced Grok 4.3 — an updated proprietary model with an improved architecture, comparable in scale to Grok 4.20, focused on efficiency and new capabilities.
May 3, 2026. NVIDIA announced Star Elastic — an innovative post-training method that integrates multiple nested reasoning models (30B, 23B, 12B parameters) into a single checkpoint for zero-shot slicing.

Sources:

LLM Stats (llm-stats.com/llm-updates)
TechCrunch: OpenAI releases GPT-5.5 Instant (techcrunch.com/2026/05/05/openai-releases-gpt-5-5-instant-a-new-default-model-for-chatgpt)
OpenAI Blog: GPT-5.5 Instant (openai.com/index/gpt-5-5-instant)

Claude

Looking for information on the latest AI model releases for the week. Great! I have good material. I'll supplement it with information on specific releases this week. Excellent! Now I have fresh information. I'll compile an overview, focusing on releases and providing facts through the lens of stability and implementation risk.

AI Week: Focus on New Model Releases (May 3–5)

The main change this week is two major updates in the text model ecosystem. Instead of a general news summary, we'll break down what has actually changed in the models, what to expect during implementation, and where the pitfalls lie.

Releases and Updates

OpenAI deployed GPT-5.5 Instant on May 5, 2026, as the new default ChatGPT model, replacing GPT-5.3 Instant, with 52.5% fewer hallucinations on critical queries and 37.3% fewer factual errors in flagged conversations. In internal evaluations, GPT-5.5 Instant produced 52.5% fewer hallucinated statements than GPT-5.3 Instant on high-risk queries in medicine, law, and finance. This is a more mundane but accurate update – responses are "more concise and to the point without loss of content" and without unnecessary emojis.
Claude Opus 4.7 by Anthropic is now generally available with a notable improvement in advanced software development, especially on the most complex tasks (announced April 16, launched in May). The model has significantly better vision: it can see higher resolution images, is more refined and creative in professional tasks, producing high-quality interfaces, slides, and documents. On an internal benchmark of 93 coding tasks, Claude Opus 4.7 increased success rates by 13% compared to Opus 4.6.
Expansion of computational power as a foundation for stability. Anthropic has signed an agreement with SpaceX to utilize the full computing power of the Colossus 1 data center, gaining access to over 300 megawatts of new capacity (over 220,000 NVIDIA GPUs) for a month. Anthropic has expanded usage limits for Claude Code and Claude API, doubling limits for Pro, Team, and Enterprise, and removing peak hour restrictions for Pro and Max.
New tools for integrating Opus 4.7. Anthropic has launched Claude Design, a new product from Anthropic Labs that allows collaboration with Claude to create visual outputs such as designs, prototypes, slides, and marketing materials. Claude Security has entered public beta for Claude Enterprise, checking code for vulnerabilities and suggesting fixes using Opus 4.7, adding scheduled scans, targeted scans, and improved triage tracking.
Technical migration risks. Claude Opus 4.7 uses a new tokenizer that may use approximately 1x to 1.35x more tokens when processing text compared to previous models (up to ~35% more depending on content). When OpenAI released the GPT-4o model, there was significant user backlash, with users describing the model as their "best friend" – new updates complicate the operation of custom prompts built on older behaviors.

Sources:

OpenAI Official Blog (GPT-5.5 Instant Release)
Anthropic Official Blog (Claude Opus 4.7 Launch)
TechCrunch, SiliconANGLE, MacRumors (model coverage)
Releasebot, GitHub Changelog (technical updates)

Sources:

Blog