Just shipped
Every notable AI platform that just launched — verified against the vendor's own announcement, with a plain-English take on what it actually does.
Latest update · Jul 21, 2026
A faster, cheaper Flash tier — plus a first Gemini 4 tease.
Google refreshed its mid-tier lineup with Gemini 3.6 Flash and 3.5 Flash-Lite: output tokens cost 17% less to use and coding benchmark scores jumped (DeepSWE 37% -> 49%), while output pricing dropped from $9 to $7.50 per million tokens. Google also confirmed Gemini 4 pre-training has begun. Worth a look if you're running high-volume, cost-sensitive Gemini workloads.
Google's new flagship model debuts at the World AI Conference.
Gemini 3.5 Pro is Google's new top-end model, launched alongside Shanghai's 2026 World Artificial Intelligence Conference. If you standardised your model mix months ago, this is the cue to re-benchmark — it reopens price and capability competition at the frontier.
ChatGPT's voice mode gets a new real-time model.
OpenAI replaced the older model behind ChatGPT's voice mode with GPT-Live, improving real-time speech quality and dropping the stale knowledge cutoff. It raises the bar for the conversational and phone-agent experiences buyers benchmark against.
No-code voice agents in under two minutes.
xAI launched a no-code Voice Agent Builder that spins up production voice agents in minutes, priced at $0.05/min of audio plus $0.01/min telephony. It adds fresh price pressure on voice-agent platforms — worth benchmarking for high-volume outbound like support and collections.
Prompt-to-video for every Google account.
Type what you want and Vids generates polished video clips with Veo 3.1, free on any Google account — plus AI avatars and custom music on paid tiers. It lives inside Workspace next to Docs and Slides, so teams can make videos together without editing skills.
OpenAI's new flagship family goes public in three tiers.
Sol handles frontier reasoning and long-horizon agents (with an 'ultra mode' that spawns subagents), Terra covers everyday work at ~2x lower cost than GPT-5.5, and Luna makes high-volume tasks cheap. If you use ChatGPT or build on the API, this is the new default lineup.
Open-weight coding model that tops agentic benchmarks.
A trillion-parameter open-weight model tuned for agentic coding — it leads Opus 4.8 on MCP-Mark Verified while costing from $0.60 per million tokens. Developers can run it via API or self-host it, making frontier-grade coding help dramatically cheaper.
Your documents, turned into narrated video explainers.
NotebookLM now turns your uploaded sources into watchable video overviews, and its new Interactive Mode lets you pause an Audio Overview to ask questions mid-stream. Great for studying, onboarding and digesting long reports.
Near-instant image generation at commodity prices.
Generates images in under four seconds from roughly $0.034 per 1,000 images, at higher quality than the original NanoBanana. Cheap, fast image generation for high-volume creative and product work.
Near-frontier coding ability at mid-tier prices.
Anthropic's new workhorse model posts 63.2% on SWE-Bench Pro at $2/$10 per million tokens — undercutting rivals on cost-per-capability. If you build with Claude or use Claude Code, Sonnet 5 is the new price-performance default.
China's answer to Cursor and Claude Code goes global.
An agentic coding assistant from Zhipu that plans and edits across whole codebases, launched globally to challenge Cursor, Claude Code and Copilot on price and speed. Expect aggressive pricing pressure across the category.
Agents that build, test and deploy full-stack apps from a prompt.
Describe the product you want and Emergent's autonomous agents plan, code, test and ship a working full-stack app — backend, database and integrations included — while you steer in plain English. One of the fastest-growing 'vibe-coding' platforms for non-coders.
The AI-native browser reaches general rollout.
A browser with an agent built in: it reads pages, runs multi-step research across tabs and takes actions on your behalf instead of just returning links. The biggest bet yet that research moves from the search box into the browser itself.
Top of the public intelligence indexes, built for agents.
Anthropic's flagship takes the lead on public intelligence indexes with much stronger long-horizon agentic execution — the model behind the current wave of autonomous coding and research agents.
Only launches we've verified against the vendor's own announcement make this page. For the wider news tape — pricing moves, funding, research and security — see the Intelligence feed.