The first week of June 2026 has already delivered a wave of new AI model releases and tools. From NVIDIA's enterprise safety model to JetBrains' coding assistant and a new breed of local computer-use agents, here's what matters.
NVIDIA Nemotron 3.5 Content Safety
Released: June 4, 2026 | By: NVIDIA
NVIDIA released Nemotron 3.5 Content Safety, a customizable multimodal safety model built for enterprise AI deployments. It checks text, images, and audio inputs/outputs in real time and lets organizations tune safety thresholds to match regional compliance requirements like GDPR, CCPA, and emerging AI-specific regulations.
Key capabilities: - Customizable safety policies per use case - Multimodal protection (text, image, audio) - Low-latency inference for real-time applications - Built-in regulatory compliance support
This fills a real gap. As companies deploy AI across customer-facing products, having a safety layer that works across modalities — not just text — is becoming table stakes.
JetBrains Mellum2: 12B MoE Coding Model
Released: June 1, 2026 | By: JetBrains
JetBrains introduced Mellum2, a 12-billion parameter Mixture-of-Experts model purpose-built for software development. Using MoE architecture means only relevant parameters activate per task, keeping inference efficient.
What makes it notable: - Trained specifically on code generation, debugging, and refactoring - Extended context windows for understanding larger codebases - Designed for IDE integration out of the box - Strong early benchmarks on code completion and bug detection
JetBrains doubling down on domain-specific AI (rather than general-purpose LLMs) is a smart play. Developers don't need a model that can write poetry — they need one that understands code structure, naming conventions, and project context.
Holo3.1: Fast Local Computer-Use Agents
Released: June 2, 2026 | By: Hcompany
Hcompany released Holo3.1, focused on running computer-use agents locally with low latency. The push toward local agent execution (rather than cloud-dependent) addresses privacy concerns and reduces the round-trip latency that makes cloud agents feel sluggish.
Why this matters: - Local execution keeps sensitive data on-device - Faster response times without cloud round-trips - Lower cost at scale (no per-token API charges) - Growing trend toward edge AI agents
ServiceNow EVA-Bench Data 2.0
Released: June 4, 2026 | By: ServiceNow-AI
ServiceNow released EVA-Bench Data 2.0, an expanded evaluation framework covering 3 domains, 121 tools, and 213 scenarios. It benchmarks AI agents on tool use, multi-step reasoning, error recovery, and resource efficiency — going beyond traditional language benchmarks to measure real-world agent performance.
This is important because the industry still lacks good ways to compare agent systems. EVA-Bench 2.0 might not be the final answer, but it's a step in the right direction.
The Bigger Picture: Claude Opus 4.8 Leads the Pack
Anthropic's Claude Opus 4.8, released May 28, currently holds the #1 spot on the Artificial Analysis Intelligence Index with a score of 61.4 — the first model to cleanly break above 60. It dethroned GPT-5.5, which had held the top spot since its April release.
The frontier model competition remains intense: - Claude Opus 4.8 — #1 on benchmarks, Anthropic's strongest model yet - GPT-5.5 — Still competitive, with Instant variant for speed-sensitive tasks - Gemini 3.1 Pro/Ultra — Google's multimodal powerhouse - DeepSeek V4 — Open-weight contender with 1M token context - Nemotron 3 Ultra — NVIDIA's 550B open model
What to Watch
The rest of June should bring more movement. Google I/O follow-up releases, potential Meta updates, and continued pressure from Chinese labs (DeepSeek, Qwen) will keep the landscape shifting. The trend toward specialized models (safety, coding, agents) over general-purpose ones is accelerating — expect more domain-specific releases in the coming weeks.