Custos LabsThe best AI tool for each job.
Our top 3 for ten everyday uses, plus notable new launches. Every pick links the leaderboard or official page it came from.
🏆 Top 3 by use case
🧑💻 Coding agents / vibe coding
- 🥇 Claude Code (Claude Opus 5.5)Anthropic
Claude Opus 5.5 (max) tops Artificial Analysis's overall intelligence index at 58, ahead of every GPT-6.x, Gemini 4 and Grok 4.7 variant listed, and Claude Code is bundled into Pro and Max.
Free tier (no Claude Code) + Pro $20/mo ($17/mo annual) with Claude Code; Max from $100/mo · source - 🥈 Codex (OpenAI GPT-6.x)OpenAI
OpenAI's own page says it runs parallel agents in the cloud, CLI and IDE and reviews PRs, and GPT-6 Astra (max) scores 53 on the AA intelligence index.
Included in ChatGPT plans (plan price not retrieved) · source
💬 General chat assistant
- 🥇 Claude (Opus 5.5 / Sonnet 5.5 / Fable 5.1)Anthropic
Claude models hold seven of the top eight spots on the AA intelligence index, and the free plan covers web search, file creation, code execution and memory.
Free tier + $20/mo Pro ($17/mo annual) · source - 🥈 ChatGPT (GPT-6 / GPT-6.1)OpenAI
GPT-6 Astra (max) scores 53 and GPT-6.1 Sol (max) scores 52 on the AA index, and the free version is open to everyone.
Free tier + paid plans Go/Plus (prices not retrieved) · source - 🥉 Gemini app (Gemini 4 Argon)Google
Gemini 4 Argon (high) scores 53 on the AA index and the free plan already includes Deep Research, Gemini Live and image generation.
Free tier + AI Plus $4.99/mo, AI Pro $19.99/mo · source
🔎 Deep research / search
- 🥇 Gemini Deep ResearchGoogle
Deep Research is on the free Gemini plan, and Google AI Pro adds Gemini Notebook with Audio Overviews as a research and writing assistant.
Free tier + $19.99/mo (AI Pro) · source - 🥈 Claude ResearchAnthropic
Claude's pricing page lists a Research feature alongside web search and connectors.
Free tier + $20/mo Pro (Research availability by tier not confirmed) · source
🎨 Image generation
- 🥇 GPT Image 2.5 (Sunburst)OpenAI
It ranks #1 on the AA text-to-image arena at 1222 Elo, ahead of the next family member at 1207 and Nano Banana 2.1 at 1176.
API $210.7 per 1k images (consumer pricing not retrieved) · source - 🥈 Nano Banana 2.1Google
It scores 1176 Elo in the top four while costing $33.6 per 1k images, roughly a sixth of GPT Image 2.5.
API $33.6 per 1k images · source - 🥉 MAI-Image-2.6Microsoft
It scores 1166 Elo (tied for 4th–6th with Grok Imagine Image 2.0) at $38.9 per 1k images.
API $38.9 per 1k images · source
🎬 Video generation
- 🥇 Wan 3.0Alibaba
It leads the AA text-to-video arena (with audio) at 1174 Elo, though its API is among the pricier at $12.00/min.
API $12.00 per min of 1080p · source - 🥈 MiniMax H3MiniMax · runs locally
It scores 1158 Elo at 768p for $4.80/min and has open weights on Hugging Face, and the H3 Max variant also ranks top 7 at 1152.
API $4.80 per min · source - 🥉 HeyGen Video 1HeyGen
#4 on Artificial Analysis' text-to-video leaderboard and the cheapest of its top models at $2.40/min; #2 Utopai X has no public API.
API $2.40 per min · source
🎙️ Voice & text-to-speech
- 🥇 Eleven v4ElevenLabs
It ranks #1 on the AA speech arena at 1334 Elo and is ElevenLabs' most emotive model; the v4 Turbo variant is statistically tied at 1333 for half the API price.
Free tier + Starter $6/mo, Creator $22/mo; API $80 per 1M chars · source - 🥈 Qwen-Audio-3.1-TTS-PlusAlibaba
It scores 1306 Elo for $19.3 per 1M characters, about a quarter of Eleven v4's API price.
API $19.3 per 1M chars · source - 🥉 Gemini 3.8 Flash TTSGoogle
It scores 1282 Elo at $16.5 per 1M characters.
API $16.5 per 1M chars · source
🎵 Music generation
- 🥇 Suno v6Suno
It leads the AA vocal-music arena at 1144 Elo, 43 points ahead of Mureka V9.5 and the largest gap on that board.
Free tier + paid plans (prices not retrievable from suno.com/pricing) · source - 🥈 Mureka V9.5Mureka
It is the only non-Suno model above 1100 Elo, at 1101.
Price not verified · source - 🥉 Lyria 3.5Google
It scores 1051 Elo, tied for 4th–6th with Mureka V9 and Lyria 3 Pro, and is the highest-placed of the big-lab models.
Price not verified · source
🧊 3D model generation
- 🥇 MeshyMeshy
Its plans include DCC bridges for Blender, Unity, Unreal, Godot and Roblox, MCP/API access and a '3D Agent', though we found no independent leaderboard so the rank is a judgment call.
Free (100 credits/mo) + Pro $20/mo · source
🖥️ Local / self-hosted models (Apple Silicon)
- 🥇 LM Studio (Bionic)LM Studio · runs locally
It ships a native macOS arm64 build running models on MLX and llama.cpp, with a local agent for work and code and on-device voice transcription.
Free (pricing not stated on page) · source - 🥈 OllamaOllama · runs locally
It claims 9M+ installs a month and runs open models locally or in the cloud without storing prompts.
Free (pricing not stated on page) · source
🎮 AI for gamers / game dev
- 🥇 Unity AI (in-editor assistant)Unity
It is a project-aware assistant inside the Editor that reads your scene, drives Editor actions and verifies changes, with a choice of models per task (benchmark claims are Unity's own).
Price not verified · source - 🥈 Meshy (game-asset 3D)Meshy
It exports to Unity, Unreal, Godot, Blender and Roblox and includes animation with 600+ motions on paid plans.
Free (100 credits/mo) + Pro $20/mo · source
✨ New AI tools
Notable launches from the last 45 days or so, newest first.
LFM3 family (1.7B text, VL-2B, Audio-2B, Omni-2B, d1)
Six ~2B-parameter models on one backbone covering text, vision with video input, audio, omni and fast decision models, designed for on-device use.
GPT-6.1 Terra
A fast, cheaper GPT-6.1 model with better visual reasoning and computer use. OpenAI reports 50.3% on Terminal-Bench 4.0 at max reasoning, against 21.5% for GPT-5.6 Terra.
Speech models in Ollama (Phonon TTS)
Ollama added text-to-speech support, starting with Gradium's Phonon models in five languages, with an OpenAI-compatible /v1/audio/speech endpoint.
Decisions API (generally available)
Endpoint built on GPT-6 Luna that returns a probability, choice or score for bounded questions about text, images or JSON, with a native GPT-Live voice integration.
Granite Speech 5.0 (2B and 470M Turbo CTC)
New 2B speech model and a multilingual 470M Turbo CTC model for transcription and translation, with 2,000-term keyword biasing; the 2B ranks #2 among open-weight models on the Open ASR Leaderboard's English average (5.06% WER).
AnyLanguageModel 1.0
Swift package that is a drop-in replacement for Apple's Foundation Models API, running local models (MLX, llama.cpp, Core ML) or hosted ones (OpenAI, Anthropic, Gemini, Ollama) with tool calling and guided generation.
Clef
27B multimodal decision model fine-tuned from Qwen3.8-27B that answers typed questions over text or images in one pass, now available on Ollama (locally or on Ollama Cloud).
Beam 1.0 (501B-A23B)
Open-weight 501B mixture-of-experts model (23B active parameters, 1M context, text-only) built for coding and agentic work, shipped with its own Rust harness.
Claude Opus 5.5
Flagship model for long-running agentic coding and knowledge work, with a 1M-token context window, 128k max output and always-on adaptive thinking.
Get the rankings when they change
We refresh the AI tool rankings regularly. Get the changes and notable new launches by email.
Free. Gumroad emails you a confirmation link first; click it to finish. Unsubscribe anytime.
How this list is made
Our AI research agent reviews public leaderboards (such as Artificial Analysis), official pricing pages and launch announcements, and we spot-check the results before publishing. Where a leaderboard exists, we follow its order. Prices can change; "Price not verified" means we couldn't confirm a current price.
Not sponsored. No one pays to be on this list and it contains no affiliate links. Disclosure: Custos Labs sells setup packs for Claude Code and Cursor, both of which appear in these rankings.