NVIDIA Nemotron 3 Ultra Sets New Open-Weight Bar

NVIDIA released Nemotron 3 Ultra, a 550B-parameter open MoE model with 55B active parameters and a 1M-token context window. The hybrid Mamba-Transformer architecture scores 48 on the Artificial Analysis Intelligence Index—the highest among US open-weight models—and delivers up to 6x higher inference throughput than comparable open LLMs.

MiniMax M3: Multimodal Coding at Frontier Level

MiniMax launched M3, the first open-weight model to combine frontier-level coding (59% on SWE-Bench Pro), a 1M-token context window, and native multimodal capabilities including desktop computer use. The company’s Sparse Attention mechanism reduces per-token compute at 1M context to roughly 1/20th of the prior generation.

Claude Opus 4.8 Advances Agent Capabilities

Anthropric released Claude Opus 4.8 with dynamic workflows in Claude Code, where the model writes orchestration scripts and spins up parallel subagents for complex tasks. The model tops ARC-AGI-3 (tripling GPT-5.5’s score), is 4x less likely to miss code flaws, and ships fast mode at 3x lower cost than prior models.

Google DeepMind’s DiffusionGemma: Text Diffusion at Speed

Google DeepMind introduced DiffusionGemma, an experimental open 26B Mixture of Experts text diffusion model with 3.8B active parameters. The model generates entire 256-token blocks simultaneously, achieving up to 4x faster inference than autoregressive models with 1000+ tokens/sec on a single NVIDIA H100. DiffusionGemma was released under an Apache 2.0 license.

Alibaba’s Qwen3.7-Plus: Affordable Multimodal Scale

Alibaba launched Qwen3.7-Plus, a multimodal agent model with 1M-token context, priced at roughly 6x less than Qwen 3.7 Max.

Google Advances Real-Time Speech Translation

Google launched Gemini 3.5 Live Translate for real-time speech-to-speech translation across 70+ languages while preserving the speaker’s intonation, pacing, and pitch.

Anthropic Releases Claude Fable 5

Anthropric released Claude Fable 5, showing major benchmark leaps at roughly 2x the cost of Opus models.

IPO Wave: OpenAI and Anthropic Move Toward Public Markets

OpenAI filed a confidential S-1 with the SEC on June 8, 2026, following a private valuation of $852 billion in March. The filing comes amid projected losses of $14 billion for the year with no expected profitability until 2029-2030.

Anthropric confidentially submitted a draft registration statement for an IPO following a $65 billion Series H round that valued the company at $965 billion.

Neura Robotics Closes Record $1.4B Series C

Neura Robotics closed a Series C funding round of up to $1.4 billion, which the company claims is the largest investment into a full-stack robotics company to date.

GitHub Copilot Shifts to Usage-Based Billing

GitHub announced it will transition Copilot from request-based billing to usage-based metered billing on June 1, 2026, introducing a virtual currency called GitHub AI Credits priced at $0.01 each.

Source: Medium (Fabio Chiusano)