← Home

Blog

Articles about AI search, verification, and platform updates

AI Digest of the Week

May 31, 2026 · 5 min read

Perplexity

Over the past week, the most notable developments were new model releases and updates from OpenAI, Anthropic, Google, and xAI, rather than platform announcements or general trends. Below is a brief overview focusing specifically on the models and their practical significance.

  1. April 21, 2026 OpenAI released ChatGPT Images 2.0 — a new image generation model that, according to digest descriptions, follows prompts significantly better, maintains composition, and confidently handles complex visual tasks.[3][6]

  2. April 16, 2026 Anthropic introduced Claude Opus 4.7; reviews present the release as an enhancement of the model in programming, agent scenarios, vision, and multi-step tasks.[3]

  3. April 2026 Google updated its Gemma 4 lineup and included new tools for generation and analysis in its monthly announcements; this release is noted in timelines as one of the key open-weight launches of the month.[3][5]

  4. April 2026 DeepSeek released DeepSeek-V4-Flash-Max and DeepSeek-V4-Pro-Max, reinforcing the trend towards fast and powerful open-source/open models in the Chinese ecosystem.[5]

  5. April 2026 Amidst major releases, the market shift towards models for code, agents, and multimodality solidified: these areas most frequently appeared in the weekly and monthly summaries as the most practical for businesses and developers.[1][2][3]

Sources:

  • Habr — "Key AI World Events in 1 Week of August 2025"
  • Sostav.ru — "New Neural Networks of the Month: April Digest"
  • Deeplist — "Latest News from the World of Neural Networks and AI"
  • AI-Stat — "Timeline of Neural Network Releases 2026"

Gemini

Major AI Model Releases of the Past Week: Focus on Stability and Autonomy

The last week of May 2026 saw the release of several significant updates in the field of artificial intelligence, with a focus on improving reliability and expanding the autonomous capabilities of models.

  1. Claude Opus 4.8 by Anthropic (May 28, 2026): The new version of Anthropic's flagship model demonstrates increased "honesty," making fewer "hallucinations" and more frequently indicating uncertainty in its responses. Opus 4.8 also features improved coding capabilities and an expanded context window of up to 1 million tokens, making it a powerful tool for complex, multi-step tasks.
  2. GPT-5.5 Instant by OpenAI (May 4, 2026): OpenAI introduced GPT-5.5 Instant as the new default model for all ChatGPT users. Improvements noted include enhanced factual accuracy, a reduction in "junk" responses (e.g., emojis and excessive formatting), and more efficient decision-making during web searches.
  3. Gemini 3.5 Flash by Google (May 18, 2026): Google released Gemini 3.5 Flash, positioning it as the fastest model in the Gemini lineup. It is designed for high-throughput tasks where low latency is critical.
  4. Grok 4.3 by xAI (May 5, 2026): The new version of Grok has received improvements in integration with X/Twitter and access to real-time data with minimal latency. Its reasoning performance is comparable to Claude Opus 4.6.

These releases highlight the trend towards creating more reliable, accurate, and autonomous AI systems capable of tackling increasingly complex tasks across various domains.

Sources:

  • TOP Academy
  • LLM Stats
  • AI-Stat
  • Last Week in AI
  • Price Per Token
  • Sostav.ru
  • 3DNews
  • Anthropic
  • Habr
  • Gazeta.ru
  • OpenAI Help Center
  • Habr
  • vc.ru
  • Focus
  • dev.by
  • BOLSHOY PROSPEKT

ChatGPT

From May 24 to May 31, 2026, several significant events occurred in the field of artificial intelligence related to new model releases:

  1. May 20, 2026: Alibaba introduced the Qwen3.7 Max model, boasting an expanded context of 1 million tokens, making it one of the most powerful Chinese models for artificial intelligence analysis.

  2. May 19, 2026: Google announced Gemini 3.5 Flash, the fastest model in the Gemini lineup, designed for rapid task execution.

  3. May 18, 2026: DeepSeek released DeepSeek-V4-Pro, a model with enhanced data processing capabilities and increased performance.

  4. May 16, 2026: xAI presented the Grok 4.3 model, designed for reasoning and multitasking tasks.

  5. May 14, 2026: Meta announced Muse Spark, a preliminary version of the model that demonstrated high results in Chatbot Arena.

Sources:

  • AI-Stat: AI Model Releases in May 2026: From Gemini 3.5 Flash to GPT-5.5
  • AI Release Tracker: Latest AI Model Releases — May 2026
  • AI-Stat: April 2026 — A Record Month for Open-Source AI
  • AI Release Tracker: Latest AI Model Releases
  • AI-Stat: AI Model Releases in May 2026: From Gemini 3.5 Flash to GPT-5.5

Grok

1. May 27 — Anthropic released Claude Opus 4.8, strengthening its position in agent and reasoning tasks through improved long context handling and reduced hallucinations in professional scenarios.

2. May 28 — OpenAI updated GPT-5.5 Instant, making the model more natural and concise in everyday communication, while also retiring older versions o3 and GPT-4.5 to accelerate user transition to current versions.

3. Week of May 24–31 — Stability AI introduced audio models capable of generating full tracks up to six minutes long, opening new horizons for creative tools and multimedia applications.

4. Week of May 24–31 — Microsoft AI's MAI-Image-2.5 entered the top 3 in text-to-image rankings based on user preference, reinforcing the ecosystem's independence from third-party models.

Sources:

  • LLM Stats (llm-stats.com/llm-updates)
  • OpenAI Help Center (help.openai.com)
  • Medium: AI Updates for the Week of 5/24/26
  • LinkedIn AI News Highlights (May 28, 2026)

Claude

Looking for fresh information on model releases from the past week. Great! There's fresh data. Let me provide an overview from a different angle — instead of a standard digest, I'll focus on the architectural shift in the releases of the past week (May 24–31).

Architecture Over Scale: How Late May Releases Are Changing the AI Game

  1. SubQ Goes Beyond Transformers (May 5) — Startup Subquadratic launched SubQ 1M-Preview with a $29M seed round, the first commercial challenge to the transformer architecture. It uses sparse sub-quadratic attention with a native 12 million token window, promising about 1/5th the cost of the frontier on long-context tasks and up to 52x attention speedup when scaling.

  2. Zyphra Breaks NVIDIA's Monopoly with ZAYA1-8B (May 6) — ZAYA1-8B was released under Apache 2.0, trained entirely on AMD Instinct, not ported but prepared from scratch on AMD. This is the first open-source leader without the NVIDIA stack — a direct signal about alternative training paths.

  3. Google Switches ChatGPT Default to New Safety Standard (May) — When OpenAI changes the default model in ChatGPT, the quality of responses for hundreds of millions of people changes overnight; OpenAI chose "fewer hallucinations on regulated topics" as the main improvement instead of "smarter" — this signals that the next round of competitive struggle isn't about GPQA.

  4. Gemini 3.5 Flash: Speed Over Scale (May 19) — Google positioned the launch around long-horizon tool-use and coding rather than chat quality; 3.5 Flash outperforms the previous frontier 3.1 Pro on "almost all" benchmarks, including coding, agent tasks, and multimodal reasoning.

  5. Prices Plummet Through Open Source (May) — DeepSeek V4-Pro, at new stable rates, is approximately 8x cheaper on input tokens and 10x cheaper on output tokens than Claude Opus 4.7, while remaining competitive on coding benchmarks.

The Main Angle: May 2026 is not a battle for +2% accuracy. It's a shift from scale to architecture, openness, and local execution. The frontier has frozen; innovation has moved down a layer.

Sources:

  • WhatLLM.org: New AI Models May 2026
  • LLM Stats: AI Updates Today
  • Codersera: AI Models Released in May 2026
  • BuildFastWithAI: Best AI Models of May 2026
  • DigitalApplied: AI Model Releases May 2026

Sources: