Every frontier AI tool, cited. Editorial-grade intel sourced from official product pages and tier-1 publications — not LLM-guessed lists or SEO farm round-ups.
§ 01 — By category
9 categories
§ 02 — The index
Photorealism, text rendering, and character consistency at Flash-tier speed
Canonical name is Gemini 3.1 Flash Image; marketed as Nano Banana 2. Pairs Nano Banana Pro capabilities with Flash speed; up to 5 characters + 14 objects preserved.
Frontier 4MP generation/editing with multi-reference identity preservation
32B latent flow-matching model pairing a Mistral-3 24B VLM with a rectified flow transformer; up to 10 reference images and JSON prompting.
No product shipped under this name — OpenAI replaced DALL-E 3 with native GPT-4o image generation in March 2025
No product named DALL-E 4 exists. OpenAI shifted image generation into GPT-4o (gpt-image-1) in March 2025; DALL-E-3 API snapshots deprecate May 12 2026.
High-fidelity photorealism, typography, and 2K creative generation
Announced at Google I/O May 20 2025. Three-tier family (Fast/Standard/Ultra) with improved typography, up to 2K output, Fast ~10x faster than Imagen 3.
Aesthetic-forward image generation with voice prompting, Draft Mode, Omni Reference
Shipped April 3 2025; default June 17 2025. Draft Mode (2x faster, 1/2 cost) and voice prompting. Reception was mixed — critics called it incremental vs V6.
Claimed image + real-time video synthesis upgrade
Only non-tier-1 outlets reference 'Stable Diffusion 4'; Stability AI's own news feed does not index an SD4 release. Stable Diffusion 3.5 remains the last confirmed version.
Bilingual (Chinese/English) text-in-image rendering for posters and e-commerce
Jan 2026 release from Z.ai. 16B hybrid: 9B autoregressive planner + 7B diffusion decoder. ~91.16% word accuracy on CVTG-2K; trained solely on Huawei Ascend.
Native multimodal T2I with long-prompt semantic understanding and text rendering
Open-sourced Sep 28 2025. 80B MoE (13B active/token); briefly held #1 T2I slot on LMArena, ahead of Nano Banana (Gemini 2.5 Flash Image).
Photorealistic 2K–4K with multi-reference and element consistency
Feb 5 2026 Kling 3.0 launch bundled Image 3.0 / Image 3.0 Omni with Video 3.0. 'O3' is Kuaishou's shorthand for the Omni tier — native 4K, multi-reference.
Sub-second photorealistic T2I with bilingual (EN/CN) text rendering
Built by Alibaba Tongyi Lab (Fal hosts it). 6B single-stream DiT, 8-step inference, sub-second 4MP at ~$0.005/MP. Top open-source T2I on Artificial Analysis.
Physical world logic with synchronized audio
OpenAI's Sora 2, launched Sep 30 2025, animates still images with improved physical accuracy and synchronized dialogue/sound; sora-2-pro targets higher-fidelity cinematic output.
Multimodal video with native audio; cinematic action consistency
Released Feb 12 2026, generates up to 15s at 1080p with native audio and strong action consistency; topped Artificial Analysis Elo leaderboard at launch.
Prompt-adherent I2V with native synchronized audio and 3D spatial sound
Veo 3.1 (Oct 2025) improves prompt adherence and realism at 1080p/24fps with natively generated dialogue, SFX, and 3D spatial audio.
Simultaneous audio-visual I2V with voice and motion control
Released Dec 3 2025; single-pass audio-visual generation for I2V at 1080p/10s, with voice control and motion upgrades.
Cinematic I2V with start/end-frame control and character reference
Dec 18 2025 Ray3 Modify brings keyframe control, character reference, and start/end-frame guidance to Dream Machine.
Character/style consistency across shots
Gen-4 (Mar 2025) uses reference images for consistent characters, locations, and styles; Gen-4.5 (Dec 2025) extended speed and V2V control modes.
V2V environment/style swap preserving motion
Sora 2's Remix applies targeted edits (setting, style, cameo) to an existing video while preserving original motion and composition.
Physics-aware V2V effects (melt, explode, squish, inflate)
Pika 2.5 iterates on Pikaffects — physics-aware V2V transformations that respect source structure; exact 2.5 launch date not confirmed in tier-1 press.
Reference-to-video with character appearance + voice preservation
Wan 2.6 (Dec 16 2025) is a 14B MoE multimodal model; up to 15s of 1080p/24fps with synchronized audio; Wan2.6-R2V preserves look and voice.
V2V style transfer and outfit/character swap preserving motion
DomoAI's V2V applies reference-image or preset styles frame-by-frame while preserving motion and audio; Subject Only mode isolates to the main subject.
Complex reasoning, agentic coding, and Artifacts (interactive in-chat apps)
Apr 16 2026 release. New xhigh reasoning level, self-verification rigor, higher-res vision, class-leading SWE-Bench Pro coding scores.
Autonomous multi-source research reports with inline citations
Agent performs dozens of searches, reads hundreds of sources, synthesizes a 5-15 page report with bolded facts and hyperlinked citations.
AI 'vibe coding' full-stack web-app builder from natural language
Stockholm-based full-stack app builder; 100k+ daily new projects; customers include Klarna and Uber. $6.6B valuation Dec 2025.
PM workspace queries, blockers/risks extraction, timeline auto-adjustment
Sept 2025 rebuild as Agents that pull context from Slack, Drive, GitHub; surface blockers, draft updates, adjust timelines across workspace databases.
AI calendar + task manager that auto-reshuffles by priority and deadlines
Dynamically timeblocks tasks, reshuffles the day when meetings run long; ASAP > deadline > due-date ranking hierarchy.
Meeting notetaker: joins calls, transcribes, summarizes, extracts action items
Records and transcribes meetings in 100+ languages with 95%+ accuracy; AI Skills and Live Assist surface action items and real-time coaching.
AI-native email client for inbox zero: triage, drafts in your voice, auto-labels
Grammarly acquired Superhuman July 2025, then rebranded entire parent as 'Superhuman' in Oct 2025. AI drafts, Auto Labels, Auto Summarize, Instant Reply.
No-code AI bots that act across 7,000+ apps for cross-tool workflows
Launched Mar 2024 as Central, evolved into Zapier Agents. Teach bots behaviors, attach Sheets/Docs/Notion as knowledge, cross-reference data across apps.
Hybrid human+AI meeting notepad: your rough notes become a clean summary
Captures device audio directly (no meeting bot), merges your rough notes with transcript into a clean follow-up summary. $1.5B valuation Mar 2026.
Agentic AI browser that executes tasks across tabs/sites
Chromium-based agentic browser launched July 2025 ($200/mo Max), made free worldwide Oct 2025; drives full browsing sessions autonomously.
Full song generation from natural language
Suno v5 (Sept 2025) delivers stereo realism and expressive vocals; v5.5 (Mar 2026) adds voice capture. Active RIAA lawsuit.
High-fidelity vocals with surgical Inpaint editing
Inpainting regenerates 2-15 second segments surgically; 48kHz stereo, stem separation. RIAA settlement with major-label licensing Oct 31 2025.
Cinematic orchestration and text/image-to-music
Google DeepMind's Lyria 3 generates cinematic text/image-prompted tracks; Pro tier extends to 3 minutes with SynthID watermarking.
Enterprise samples, SFX, foley, 3-min ambient beds
Current flagship is Stable Audio 2.5 (Sept 2025), not 3.0. Enterprise: 3-min tracks in <2s, inpainting, licensed training data.
Copyright-cleared commercial tracks via API
Eleven Music (Aug 2025) ships cleared-for-commercial-use songs on Merlin/Kobalt-licensed catalogs; iOS app and Music Marketplace Apr 2026.
Niche genre adherence, multi-style vocals
MiniMax Music 2.0 (Oct 2025) delivers genre-faithful vocals across 16 styles × 11 emotions; v2.6 (Apr 2026) adds AI Cover and Lyrics Optimizer.
MIDI export and stem-level editing for DAW workflows
Mureka V8 (Jan 2026) from Kunlun Tech offers MIDI export, stem separation, foley editing; DAW-centric producers starting material.
Free viral short-form clips for TikTok/social
YC W24. Unlimited free AI songs with commercial rights; V3 Preview Melodia diffusion model streams in ~15s. Ideal for TikTok clips.
Classical and game scoring with MIDI output
AIVA composes orchestral/cinematic scores with MIDI/WAV export; SACEM-registered. Pro tier grants full copyright ownership. '3.0' branding unverified.
Royalty-free tracks with mood/genre sliders
Creators pick genre/mood/length sliders for royalty-free tracks up to 10 min; lifetime license on generated tracks, Premiere plugin supported.
Industry-standard cloning and expressive TTS
Alpha launched June 2025. 70+ languages, inline audio tags ([excited], [whispers]), multi-speaker dialogue. Non-realtime — use v2.5 Turbo/Flash for live agents.
Film-grade speech-to-speech performance voice swap
Ukrainian studio behind Darth Vader/Luke Skywalker recreations. Speech-to-speech preserves performance acting; explicit consent-based ethics framework.
Low-latency real-time streaming synthesis for NPCs/CS agents
Alibaba streaming TTS with ~150ms first-packet latency. Unified streaming/offline LLM backbone with FSQ and chunk-aware flow matching.
Marketing, e-learning, and explainer voiceover at scale
200+ voices across 20+ languages, drag-drop video timeline. Nov 2025 Falcon API: 55ms model latency, ~$0.01/min, SOC2/ISO 42001 certified.
Emotion-tagged voice generation and cloning
Chatterbox Turbo supports paralinguistic tags ([cough], [laugh]) plus emotion-exaggeration slider. Every output carries Perth neural watermark.
Fast STT (primary) plus developer TTS API
Nova-3 claims 54% WER reduction in streaming, <300ms latency, 10-language real-time codeswitching. First voice AI with self-serve vocabulary customization.
3-second zero-shot voice clone benchmark
First TTS paper to claim human parity on LibriSpeech/VCTK from 3-sec prompt. Microsoft explicitly declines productization citing deepfake risk.
Real-time voice mask for Discord, Twitch, gaming
Speech-to-speech converter preserving emotion/pacing. Raised $6M in 2023; 480k+ users. Targets gamers, VTubers, streamers.
Efficient multilingual OSS TTS for edge deployment
S1 (4B) + S1-mini (0.5B) support 80+ languages, 10-30s reference cloning, prosody tags. Weights non-commercial; tops Seed-TTS WER leaderboard.
Convert PDFs, articles, and books into AI-narrated audio
1,000+ voices, 60+ languages. Won Apple Design Award 2025 for Inclusivity; 50M+ users. Built by dyslexic founder Cliff Weitzman for accessibility.
The "gold standard" AI-first IDE
Anysphere's VS Code fork hit $2B ARR by Feb 2026 on Series D at $29.3B; Cursor 3 pushes agents as the default surface.
Terminal-first agentic coding and multi-file refactor
Anthropic's agentic CLI; 2.0 added checkpoints, native VS Code extension, redesigned desktop app with parallel sessions.
Enterprise monorepos with agentic Cascade flow
Codeium's AI IDE, acquired by Cognition (Devin) for ~$250M in Dec 2025; Cascade orchestrates multi-file edits across 4,000+ enterprise accounts.
Asynchronous team PR reviews and automated CI fixes
Announced at Build 2025, GA by late 2025; opens draft PRs, runs in GitHub Actions sandboxes, closes the loop with agentic code review on 60M+ reviews.
Fully autonomous ticket-to-PR agent
Cognition's autonomous engineer; 13.86% SWE-bench at launch (HN contested the diffs); grew $1M→$73M ARR in 10 months and acquired Windsurf.
Privacy-first OSS agent with BYO-model and local LLM support
Rebranded from Claude Dev; AGPL agent with file/terminal/browser tools, 15k+ stars, connectors for Claude, GPT, Gemini, Bedrock, Ollama, LM Studio.
Spec-driven cloud-native development with agent hooks
AWS's July 2025 agentic IDE generates specs, design docs, and task lists from prompts; GA just before re:Invent 2025 with MCP and event-driven hooks.
Zero-latency Rust-native editor with agentic AI panel
GPU-rendered Rust editor from Atom co-creator Nathan Sobo; Agent Panel and Zeta2 autocomplete run at 120fps with native multi-file agentic edits.
Free AI-first learning IDE for students and solo devs
ByteDance's Jan 2025 VS Code fork offers free frontier models; 6M+ registered users, 1.6M MAU, but security researchers flagged extensive telemetry.
AI test suite generation and code-integrity review
Codium AI rebranded to Qodo Nov 2024; Series A $50M early 2025 funds Qodo Cover, an autonomous regression-testing agent.
High-fidelity sculpting with PBR
Jan 2026 Preview: refined organic geometry, PBR (diffuse/roughness/metallic/normal), up to 600K faces, A/T-pose characters. Commercial license is premium-plan only.
Fast artistic 3D concepts (~10s)
Text-to-3D foundation model generating quad-retopo meshes with PBR materials in ~10s. Launched Jan 2024 with Luma's a16z-led Series B.
Geometric accuracy for 3D printing & gamedev
3D-DiT hierarchical carving at 1536³ (~3.6B voxels), 3x precision vs prior. Global site launched Nov 26 2025.
Rapid previews (~50-90s)
Tripo's fastest image/text-to-3D tier (~50-90s), not sub-5s as some claims suggest. Good for placeholders, reduced fine-detail.
Pro-grade clean topology with multi-model backend
Aggregator outputting retopologized PBR meshes for Unity/Unreal/Three.js with automated retopology, UV unwrap and multi-view generation.
VFX character swap from video (markerless mocap alt)
Acquired by Autodesk May 21 2024. Wonder Studio tracks actor performance from footage and retargets to CG characters.
2D concept art → production game meshes (AI+human)
Hybrid pipeline: AI base mesh then 80+ in-house artists clean topology, UVs, textures. Hard-surface props/vehicles strongest.
ArchViz from floor plans (2D rendered output only)
Converts 2D plans into photoreal renders with AI furniture placement; output is 2D imagery, not exportable 3D geometry.
AI image-to-material textures
Sampler AI Image-to-Material generates PBR channels (base color, roughness, normal, height, AO) from a single photo; Firefly adds text-to-texture.
Simulation-ready industrial assets for robotics
ByteDance Seed DiT producing watertight manifold meshes with PBR, built for robotics/embodied-AI physics sims (Isaac Sim, Omniverse).
Conversational answer engine with cited web sources and deep research
Most litigated answer engine of the cycle: NYT sued Dec 5 2025 alleging 175k scraping attempts in Aug 2025 + robots.txt circumvention, joining News Corp, Chicago Tribune, Britannica, Merriam-Webster, Nikkei, Asahi Shimbun, Reddit. Perplexity moved to dismiss Feb 27 2026.
Conversational multi-step AI search inside Google Search with source links
AI Mode expanded to 180 countries Aug 21 2025; Gemini 3 Flash became default in Gemini app and AI Mode Dec 17 2025. AI Overviews reach 1B+ monthly users; AI Overviews and AI Mode began merging Dec 2025.
Agentic AI browser and workspace with Autopilot Mode for knowledge work
$275M Series B Nov 20 2025 at $1.25B valuation (unicorn status). Founded by ex-Baidu execs Eric Jing + Kay Zhu; crossed $50M ARR within five months of launch.
AI answer engine over 220M+ peer-reviewed papers with evidence-weighted Consensus Meter
8M+ users searching 220M+ peer-reviewed papers. Scholar Agent built on GPT-5 + Responses API orchestrates Planning/Search/Reading/Analysis sub-agents; adopted institutionally by UNE, St. Thomas, Oregon State.
Privacy-first ad-free visual AI search with card-style conversational results
Founded 2021 by Angela Hoover + Jed White (YC W22). Logs no queries, stores no cookies, blocks FLoC tracking. PCMag Best Free AI Search Engine for 3 straight years (2024-26); Talc AI SearchBench reports 87% accuracy.
Anonymous chatbot access to frontier models without tracking or training use
Exited beta Mar 6 2025; rebranded to Duck.ai with anonymous access to GPT-4o mini, o3-mini, Claude 3 Haiku, Llama 3.3, Mistral Small 3. Provider contracts bar training-use; chats deleted within 30 days.
AI search engine for developers — discontinued Jan 16 2026
Ceased operations Jan 16 2026 roughly a month after a reported $10M raise. Not acquired — voluntary wind-down as foundation models (OpenAI, Google, Anthropic) absorbed web search and collapsed Phind's moat. Search interest already down ~91% from peak.
Search engine for AI agents — neural retrieval over a proprietary web index via API
$85M Series B Sep 3 2025 at $700M valuation, led by Benchmark with Lightspeed, YC, and NVIDIA's NVentures (Peter Fenton joined board). Funds a 5x GPU scale-up for the index powering agent retrieval — infrastructure, not consumer Perplexity clone.
Source-grounded AI notebook reasoning only over user-uploaded docs, now with web Deep Research
Nov 13 2025 added Deep Research — an agent that builds a research plan, browses hundreds of sites, returns a grounded report into the notebook. Same release expanded ingest to Sheets, .docx, images, PDFs/Drive URLs; Fast and Deep modes compose with Audio/Video Overviews.
Pivoted from consumer AI search to enterprise agentic operations (Komo Playbook)
Repositioned from consumer AI search to enterprise agentic ops; Komo Playbook turns typed or video-recorded SOPs into 24/7 autonomous workflows. PitchBook lists ~$2.98M total raised; corporate structure thin in public coverage — subscribers should treat as in-flux.
§ 03 — Editorial note
Most AI tool lists are SEO filler, hallucinated by chatbots, or stale within a month. The Atlas is researched by hand, with every claim tied to a source you can read.
Tools flagged Unverified are where we couldn't confirm the marketing — that's a signal, not a bug. Shutdowns, lawsuits, and quiet pivots get surfaced because that's what subscribers actually need to know.