Apps.

Commercial-safe, trained on licensed data.

Reportedly in development, code-named N50.

Beam AI, plus the finance specialist Hebbia.

Glasses with a colour microOLED display and the Noa AI assistant with long-term memory.

Newer entrants to the computer-use space.

Browses for the longest, most structured reports.

Codex-powered work agents; cloud-only, on Business and Enterprise plans.

Powered by Opus 4.8 — SWE-bench Pro 69.2 %, Verified 88.6 %. Dynamic workflows with parallel subagents; fast mode 3x cheaper.

Desktop agent now with Claude for Small Business: 15 agentic workflows across finance, ops, sales, marketing, HR and CS. Opus 4.8 upgrades flow through.

SWE-bench Pro 69.2 %, Verified 88.6 %; dynamic workflows enable hundreds of parallel subagents. Anthropic filed a confidential S-1 at ~$965 B valuation.

The leading open-source coding agents.

Claude's screen-driving capability.

Enterprise multi-team management GA; new pricing at $32–$96/seat/mo. SpaceX plans $60 B acquisition post-IPO (expected close July).

Open-weight model with a memory-efficient architecture.

$1 B+ Series D at $26 B valuation; ARR $492 M (1,230 % YoY). Enterprise customers include Goldman Sachs, Citi, Mercedes-Benz and NASA.

Formerly Windsurf — rebranded to Devin Desktop (OTA update June 2). Devin Local rewritten in Rust; Agent Client Protocol (ACP) support. ARR $492 M.

Crossed $500 M ARR; Music v2 model with better vocals/instrumentation and inpainting, API pricing cut up to 50 %. Series D extension adds BlackRock, NVIDIA, Wellington.

Glasses with a minimalist display and no camera.

Best free tier.

Best for global teams.

The open-source frontrunner; competitive at much lower cost.

Oura Ring 5 pre-orders open, shipping June 4 ($399-$499) with Health Radar and GLP-1 tracking. Oura filed confidentially for IPO (~$11 B valuation). Ultrahuman Ring Pro launched in US ($479).

Leads most published reasoning benchmarks and has the cheapest output among the majors; paired with Gemini Spark for long cloud tasks.

Built for long async research workflows.

An all-in-one work agent with a mixture-of-agents design that can even make phone calls.

Usage-based billing live — all plans on GitHub AI Credits. Opus models removed from $10 Pro. MAI-Code-1 now rolling out as a model option.

Briefly held #1 on a hard coding benchmark, the week closed models lost their leaderboard lock.

Multi-agent release with dynamic sub-agents and scheduled background tasks.

GPT-5.5 Instant update — cleaner formatting, reduced bullet-heavy output. Memory/personalisation expanding to Plus/Pro. GPT-4.5 sunset announced for June 27.

ChatGPT Images 2.0; best for text and infographics, and the replacement for the retired DALL-E 2 and 3.

The bot-free favourite of VCs and consultants; captures audio locally on a Mac with no bot in the participant list.

Grok Build 0.1 entered public API beta — dedicated coding model, 256 K context. Grok Connectors launched (SharePoint, Outlook, Google Workspace, Notion, GitHub, Linear, custom MCP). xAI-Cloudflare partnership confirmed.

CLI coding agent with up to 8 parallel sub-agents and Arena Mode; local-first, no source code transmitted.

The only one pulling live data from X, so it wins on breaking news.

#1 on image-to-video arena at launch; native synced audio, 720p/24fps.

Currently tops the Artificial Analysis leaderboard.

Avatar and talking-head video.

Open-weight release (9.3 B params, Apache 2.0 code). Single-stream Diffusion Transformer with native 2K, transparent backgrounds and structured JSON captions. Tops open-weight DesignArena leaderboard.

Open-weight model whose Agent Swarm scales to 300 parallel sub-agents.

The cheapest premium model; excellent at multi-shot cinematic sequences.

Meta acquired Limitless (Dec 2025); Amazon acquired Bee (mid-2025).

Microsoft's first in-house reasoning model — 35 B active / ~1 T MoE, 256 K context, zero OpenAI data.

A Chinese-origin general agent that runs long multi-step jobs end-to-end with minimal input in a cloud sandbox.

Desktop screen-driving from Manus.

Agentic automation inside Excel, governed by Agent 365; the easiest fit if you live in Microsoft 365.

Built into the meeting platforms themselves.

Always-on personal agent across M365 — proactive meeting prep, scheduling, routine tasks.

Still the king of artistic and stylised output.

Open-weight model combining frontier coding, 1 M context, and native multimodality with 20x less compute at 1 M context.

Meta's proprietary frontier model (launched Apr 8). Replaces the Llama open-weight line as Meta's top offering.

Gemini 3 Pro Image; leads the image-arena leaderboard, with the best multilingual text rendering, free in the Gemini app.

The OpenClaw ecosystem: lighter-weight (Nano/Zero), security-hardened (Iron), self-improving with memory (Hermes), NVIDIA's enterprise take (Nemo), and multi-agent coordination (Hi).

Both entered the meeting-notes space in 2025.

An always-on pendant.

Codex Sites launched (preview) — deploy web apps directly from Codex. 6 new business plugins (sales, data, creative, design, equity, IB). Expanding into ChatGPT app.

The self-hosted personal agent that started the category; runs 24/7 through a central controller.

Sunsetted as a standalone product — merged into ChatGPT agent mode and the Atlas browser (July 2025). OSWorld 38.1 %.

Live captions and conversational search.

Hybrid AI computing platform announced (with Intel) — auto-distributes between on-device and cloud models. Tripled annualised revenue since start of year. Comet browser now on all platforms.

Exited waitlist — now available to all Mac users (Perplexity Max, $200/mo). Runs 24/7 on Mac mini, controllable from iPhone.

Pika, Luma Dream Machine / Ray3, and Hailuo (MiniMax).

An always-on pendant; won Forbes' 2026 Best AI Wearable.

Resemble AI added Speech-to-Speech and deepfake detection. Descript launched API in open beta with reasoning models and AI video gen from text prompts.

Shut down May 4, 2026. Visual-processing approach was compute-heavy and error-prone; tech folded into Gemini Agent and Chrome auto-browse.

Multimodal (image + video), deep reasoning, 1 M context, 60 % cheaper than prior Qwen3.7-Max. Proprietary (API-only).

Leaked internal memo reveals 4 new glasses models (Modelo, Luna, RBM2 Refresh, Mojito VIP) plus an AI pendant, targeting 10 M wearable units. Wearables for Work business subscription planned.

V4.1 shipped May 14 — cleaner photorealism, better object understanding, 3D rendering. Highest-ranked text-to-image model outside Google and OpenAI on Artificial Analysis.

Director controls and storyboards.

Currently tops the Artificial Analysis leaderboard.

OpenAI stepped back from the consumer video race.

$400 M Series D at $5.4 B valuation; v5.5 is current with voice cloning and Studio DAW. Record labels allege 61,000+ additional unlicensed songs.

Best raw fidelity and stem separation, with a cleaner licensing story after Universal settled.

An on-device assistant with its own memory and credentials the model can never read (credential isolation).

New Ingredients to Video feature (generate from reference images, native vertical output). Veo 3.1 Lite at $0.05/sec (< 50 % of Veo 3.1 Fast). Free generation for all Google accounts.