AI models
Every AI app runs on a model, and new ones arrive most weeks. These pages say what each model is for, when it came out, which app or plan gives access to it, and what it costs through the API. Facts come from the maker's own announcement or docs and are dated.
Updated monthly.
More tools
GPT-5.1-Codex-Max
OpenAI coding model built for long agent sessions, now marked deprecated in the API
GPT-5.2
OpenAI flagship from December 2025, now an older model kept in the API
GPT-5.4
OpenAI model from March 2026 with built-in computer use and a 1M-token context
GPT-5.4 mini and nano
Two smaller, cheaper OpenAI models for high-volume jobs and coding sub-agents
ChatGPT Images 2.0
OpenAI's April 2026 image model, better at dense text and world knowledge, with a new thinking mode
GPT-5.5
OpenAI frontier model from April 2026 for coding, computer use and research work
GPT-Live
OpenAI voice model that can listen and speak at once, behind ChatGPT Voice
GPT-6 Astra
OpenAI top model as of September 2026, built for long tasks from request to finished result
ChatGPT Images 2.5
OpenAI's September 2026 image model with sharper detail, steadier edits and up to 50% faster generation
Astra for Law
OpenAI's GPT-6 Astra set up with a US legal search index for legal research and drafting
GPT-6 Luna
Cheap, fast OpenAI model from the GPT-6 family for high-volume and focused tasks
GPT-6 Sol
Mid-priced OpenAI GPT-6 model for complex coding and agent workflows
GPT-6.1 Sol
OpenAI model close to GPT-6 Astra in quality at a much lower price
Claude Opus 4.5
Anthropic's November 2025 flagship model, aimed at coding, agents and computer use, at a lower Opus price
Claude Opus 4.6
Anthropic's February 2026 Opus upgrade, the first Opus model with a 1 million token context window
Claude Sonnet 4.6
Anthropic's February 2026 mid-sized model, which became the default for Free and Pro users
Claude Opus 4.7
Anthropic's April 2026 Opus model, stronger on hard coding work and with sharper image vision
Claude Opus 4.8
Anthropic's May 2026 Opus update, more reliable on agent work, with a cheaper fast mode
Claude Fable 5
Anthropic's June 2026 Mythos-class model, released with safeguards so it could be offered to everyone
Claude Sonnet 5
Anthropic's June 2026 Sonnet model, close to Opus 4.8 on agent and coding work at a lower price
Claude Fable 5.1
Anthropic's September 2026 top model for coding and knowledge work, with much cheaper cache reads than Fable 5
Claude Opus 5.5
Anthropic's September 2026 Opus model, close to Fable 5.1 on most work and cheaper to run than Opus 5
Cursor Composer 1
First in-house Cursor coding model, built for fast agent work and released with Cursor 2.0
Cursor Composer 1.5
Second Cursor coding model, a thinking model with more training and self-summarising for long tasks
Cursor Composer 2
Cursor coding model from March 2026 that added continued pretraining and cut the price per token
Cursor Composer 2.5
Current Cursor in-house coding model, tuned for long agent sessions and more reliable behaviour
DeepSeek-OCR 2
Open DeepSeek model that reads documents and images and turns them into text or Markdown
DeepSeek V4.1 Flash
DeepSeek's efficient open-weight model that reads images and text, with a one million token context
Seedream 5.0 Pro
ByteDance's image model for infographics, precise region edits and multilingual text
Seedance 2.5
ByteDance's video model that makes up to 30 seconds of video with sound in one generation
Seedance 2.0
ByteDance's video model that makes clips with sound from text, image, audio and video references
Seedream 5.0 Lite
ByteDance's image model that reasons over prompts and can search the web before drawing
Seedream 4.5
ByteDance's image model with stronger multi-image editing, typography and dense text
Eleven v3
Expressive ElevenLabs text-to-speech model with audio tags and multi-speaker dialogue
Scribe v2 Realtime
ElevenLabs' live transcription model with about 150 millisecond latency for voice agents and meetings
Scribe v2
ElevenLabs' speech-to-text model for accurate batch transcription, subtitles and captions in 90+ languages
Eleven v4
ElevenLabs' most emotive text-to-speech model, with a fast Turbo version for voice agents
FLUX.2 [klein]
Small, fast FLUX.2 image models in 4B and 9B sizes that generate and edit in about a second.
FLUX 3 Image
Black Forest Labs image model that places each element where a drawn box says, with 4K output and precise edits.
GLM-5.3-Flash
Low-cost Z.ai model that sees images and video and checks its own visual output
Veo 3.1
Google's video generation model with generated sound, used in Flow, the Gemini app and the Gemini API
Veo 3.1 Lite
The cheapest model in Google's Veo 3.1 video family, made for developers generating video in volume
Lyria 3.5
Google DeepMind's newest music model, with more realistic vocals, better lyrics and songs up to three minutes
Gemini 2.5 Flash Image (Nano Banana)
Google's 2025 image generation and editing model, known as Nano Banana, now being retired
Gemini 3
Google's third generation of Gemini models, launched November 2025 with Gemini 3 Pro
Nano Banana Pro
Google's highest-quality Nano Banana image model, built on Gemini 3 Pro, for detailed images and legible text
Gemini 3 Flash
Faster, cheaper Gemini 3 model released December 2025 for everyday tasks and agents
Gemini 3.1 Deep Think
Gemini's slowest, most thorough reasoning mode for science, maths and engineering problems
Lyria 3
Google DeepMind's music model that makes 30-second songs with lyrics from a text prompt, photo or video
Gemini 3.1 Pro
Google's upgraded Pro model from February 2026 for complex reasoning and coding
Nano Banana 2
Google's fast image generation and editing model that brings Nano Banana Pro features to Flash speed
Gemini 3.1 Flash-Lite
Google's cheapest, fastest Gemini 3 model for high-volume developer work
Gemini Embedding 2
Google's first embedding model that maps text, images, video, audio and PDFs into one space
Gemini 3.1 Flash Live
Google's March 2026 real-time voice model for spoken conversations and voice agents
Gemini 3.1 Flash TTS
Text-to-speech model from April 2026 with inline audio tags to direct voice style
Gemini 3.5 Flash
Google's May 2026 Flash model aimed at agents and coding, now replaced by newer Flash versions
Gemini Omni Flash
Google's video model that makes and edits clips from text, images, audio or video
Nano Banana 2 Lite
Google's fastest and cheapest Nano Banana image model, producing a picture in about four seconds
Gemini 3.7 Flash
August 2026 Flash model for coding and agents, since superseded by Gemini 3.8 Flash
Gemini 3.8 Flash
Google's September 2026 Flash model for long coding tasks, agents and business workflows
Gemini 3.8 Live
Google's September 2026 voice models that can reason while holding a spoken conversation
Gemini 4 Argon
Google's next frontier model, so far released only to vetted cyber defenders and testers
Gemini 3.5 Transcribe
Google's speech-to-text model with speaker labels, timestamps and custom vocabulary
Gemini 3.5 Live Translate
Google's live speech-to-speech translation model covering more than 70 languages
Grok Voice Think Fast 1.0
xAI's voice model that reasons in the background during calls, aimed at support and sales
Grok 4.3
Low-cost Grok model with a 1M-token context and strong tool calling
Grok 4.5
SpaceXAI coding and agent model trained alongside Cursor, priced below rival flagships
Grok Voice Think Fast 2.0
xAI's second voice model that reasons while it speaks, with better transcription and tool use
Grok Imagine Image 2.0
xAI's image model for generating and editing pictures, with sharper text and layout
Grok 4.6
SpaceXAI model from August 2026 aimed at long-running agents and app building
Hailuo 2.3
MiniMax's video model with better body movement, facial expression and art styles
MiniMax H3 Max
fal's retrained version of MiniMax's H3 video model, tuned for prompt following and speed
Avatar V
HeyGen's avatar model that builds a consistent digital twin of a person from a 15-second recording
Higgsfield Soul
Higgsfield's first fashion-focused photo model, with one-tap style presets
Higgsfield Soul 2.0
Higgsfield's second photo model, built for fashion-aware images with consistent characters
Holo3.1
H Company's computer-use models in small sizes that can run fully on local hardware
Ideogram 4.0
Ideogram's first open-weight text-to-image model, built for accurate text and layout in images
Ideogram 4.5
Ideogram image model aimed at editing, built to keep pictures clean across many rounds of changes
Mercury 2.5 (Inception)
Inception's larger diffusion language model, faster and cheaper than most small frontier models, with a 260K context
Inkling-Small
Smaller open Inkling model with 12 billion active parameters that nearly matches the full version at lower cost
Inworld Realtime TTS-2
Inworld's text-to-speech model for live voice agents that adjusts its delivery to the caller's tone
Kimi K2
Moonshot AI's first trillion-parameter open-weight model, built for coding and tool use
Kimi K2 Thinking
Reasoning version of Kimi K2 that thinks step by step between hundreds of tool calls
Kimi K2.5
Moonshot AI's open multimodal model that added image and video input and agent swarms to Kimi
Kimi K2.6
Open Moonshot AI model for long coding sessions and swarms of up to 300 sub-agents
Kimi K2.7 Code
Moonshot AI's open coding model, built on K2.6 and using about 30% fewer thinking tokens
Kimi K3
Moonshot AI's 2.8 trillion parameter open-weight flagship with a million-token context and vision
Kling O1
Kling video model that handles generation and text-prompted editing of video in one system
Kling 3.0
Kling's video model with clips up to 15 seconds, multi-shot scenes and audio in five languages
Kling 2.6
Kling's first video model to generate speech, sound effects and ambience together with the picture
Krea Realtime 14B
Open video model from Krea that streams frames live, so prompts can change while a video plays
Krea 2
Krea's first in-house image model, built for style control with reference images and moodboards
Mistral 3
Mistral's December 2025 family of open models, from the small Ministral 3 sizes up to Mistral Large 3
Devstral 2
Mistral's open-weight coding model family, now retired from the API in favour of Mistral Medium 3.5
Mistral OCR 3
Mistral's December 2025 document reading model that pulls text, tables and images out of PDFs and scans
Voxtral Transcribe 2
Mistral's speech-to-text models for batch and live transcription in 13 languages, one with open weights
Mistral Small 4
An open Mistral model that handles chat, reasoning, coding and images in one 119B mixture-of-experts model
Voxtral TTS
Mistral's first text-to-speech model, with expressive voices in 9 languages and voice cloning
Mistral OCR 4.1
Mistral's current document reading model, returning text with layout boxes, block types and confidence scores
LTX-2
Lightricks' 2025 open-source video model that first added synchronised sound and 4K output
LTX-2.3
Lightricks' 2026 open video model with sound, sharper detail and native vertical video, runnable on home GPUs
LTX-2.5
Lightricks' latest open-weight video model, with multi-shot scenes, sound, 4K output and HDR
LTXV (LTX-Video)
First generation of Lightricks' open-source LTX video model, fast enough for near real-time output
Ray3.14
Luma video model with native 1080p output, faster and cheaper than the original Ray3
Uni-1
Luma's image model that reasons about a request in language before generating or editing a picture
Ray3.2
Luma's video model update with up to 16 keyframes per clip and HDR output for production work
Muse Spark
The first model from Meta Superintelligence Labs, a multimodal reasoning model behind the Meta AI app
Muse Spark 1.1
Meta's July 2026 update to Muse Spark with a 1M-token context, computer use and the first public API
Muse Glimmer
Meta's open-weight 30B model for running AI agents locally on a Mac or PC with one consumer GPU
Muse Voice Transcribe
Meta's real-time speech-to-text model that labels 20+ speakers and follows language switches mid-sentence
Muse Spark 1.3
Meta's September 2026 Muse Spark update for long agent tasks and coding, with a max reasoning setting
Muse Image
Meta's in-house image model that plans a layout before drawing, used in Meta AI and the Meta Model API
Meta Llama models
Meta's family of downloadable large language models, from Llama 2 to the multimodal Llama 4
Microsoft Mu
A small Microsoft language model that runs on the NPU of Copilot+ PCs and powers the Windows Settings agent
MAI-Image-1
Microsoft's first in-house image generation model, added to Bing Image Creator in late 2025
MAI-Image-2
Microsoft's second in-house image model, with better photorealism and readable text inside images
Nemotron 3 Super
NVIDIA's open 120B reasoning model with a 1 million token context, built for multi-agent systems
Nemotron 3 Nano Omni
An open NVIDIA model that reads video, audio, images and text together for AI agents
NVIDIA Nemotron 3 Ultra
NVIDIA's largest open-weight reasoning model, built for AI agents, coding and long documents
Nemotron 3.5 Lightning
A small, fast open NVIDIA model that handles the routine tool calls in long-running AI agents
Pika Audio
Family of four Pika sound models for soundtracks, music, sound effects and speech, sold through its API
Qwen VLo
Preview multimodal model from Alibaba that understands images and generates or edits them from chat prompts
Qwen3-Coder
Alibaba's open-weight coding model built for agent tasks, led by a 480B mixture-of-experts version
Qwen-Image-Layered
Open Alibaba model that splits an image into separate transparent layers so each part can be edited alone
MAI-UI
Alibaba Tongyi research models that operate smartphone apps by looking at the screen and tapping
Qwen3-TTS
Open-source Alibaba text-to-speech models with voice cloning from 3 seconds of audio and voice design
Qwen3-Max-Thinking
Alibaba's flagship reasoning model, available in Qwen Chat and through the Alibaba Cloud API
Qwen3-Coder-Next
Small, efficient open-weight Qwen model for coding agents, with 3B active of 80B parameters
Qwen3.5 Small
The small open models in Alibaba's Qwen3.5 series, from 0.8B to 9B parameters, for phones and laptops
Qwen3.6-27B
First open-weight Qwen3.6 model, a 27B model with vision focused on agentic coding
Qwen3.7-Max
Alibaba's largest Qwen3.7 model, a closed text model aimed at agents, coding and long tasks
Qwen-Image-3.0
Alibaba's hosted image model built for text-heavy layouts, taking prompts up to 4,500 tokens
Qwen-Image-2.0
Alibaba's 7-billion-parameter model that both creates and edits images, with strong text rendering at 2K
Qwen3.5-Omni
Alibaba model that takes text, images, audio and video together and replies in text or speech
Recraft V4
Recraft's rebuilt image model tuned for design work, with editable vector output
Recraft V4.1
An update to Recraft's image model with more natural photos, new illustration styles and a simpler Utility variant
Reve 2.0
Reve's June 2026 image model that builds each picture from an editable layout in native 4K
Reve 2.1
Reve's native-4K image model with element-level editing and better text in other scripts
Runway Aleph
Runway's video model that edits existing footage from text, adding or removing objects and changing angles or light
Runway Gen-4.5
Runway's video generation model for realistic, cinematic clips from text and images
Suno v6
Suno's sixth-generation song models, built with Warner Music Group, BMG and Believe
Tavus Phoenix-4
Tavus model that renders a lifelike talking face in real time, with expressions that react to the conversation
Canto
Wispr Flow's speech recognition model built for dictation in noisy, real-world conditions
FLUX.1 Kontext [dev]
Open-weight image editing model from Black Forest Labs that edits pictures from text instructions on consumer hardware.
Chai-2
Chai Discovery's AI models for designing new antibodies from scratch, in early access
DeepSWE-Preview
Open-source coding agent model trained with reinforcement learning on top of Qwen3-32B
Hunyuan3D-PolyGen
Tencent's 3D model that rebuilds AI-generated meshes with clean, game-ready polygon layouts
VibeVoice
Microsoft's family of open-source research models for long-form speech generation and transcription
Wan2.2-S2V
Alibaba's open model that turns one picture and an audio clip into a video of the person talking or singing.
Stable Audio 2.5
Stability AI's music and sound model aimed at brands and studios making commercial audio
HunyuanImage 3.0
Tencent's large open-weight image model that writes and edits pictures from text
SWE-grep
Cognition's fast model that searches a codebase in parallel to find the code an agent needs
PokeeResearch 7B
Open-source 7-billion-parameter model from Pokee AI built to run multi-step web research and cite sources
HunyuanWorld-Mirror
Tencent's open model that rebuilds 3D scenes from photos or video in a single pass
Odyssey-2
Odyssey's world model that streams AI video in real time and changes it as people type prompts
ERNIE 5.0
Baidu flagship AI model that reads and generates text, images, audio and video within one system.
Holo2
H Company's small computer-use models that click and type their way through web, desktop and phone apps
SAM 3D
Meta's pair of open models that turn a single photo into 3D objects or 3D human bodies
Olmo 3
Ai2's fully open family of 7B and 32B language models, released with training data and code
FLUX.2
Black Forest Labs image model family for photoreal images, readable text and editing with up to 10 references.
HunyuanOCR
Tencent's small open model that reads text in documents and images and can translate it
Lux
OpenAGI's computer-use model that operates desktop and web software by reading screenshots and taking actions
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Rnj-1
Essential AI's open 8-billion-parameter language model aimed at coding, tool use and maths
GLM-4.6V
Open-source Z.ai vision model that reads images and video and can call tools directly
Molmo
Ai2's family of fully open vision models that describe, point to and track things in images and video
NVIDIA Nemotron 3
NVIDIA's family of open mixture-of-experts models in Nano, Super and Ultra sizes for building AI agents
SAM Audio
Meta's open model that pulls one sound out of a recording or video from a text, click or time prompt
Wan 2.6
Alibaba's December 2025 video and image models, including a mode that puts a real person into new scenes.
GLM-4.7
Z.ai open-weights model from late 2025 focused on coding, reasoning and agent tasks
MiniMax M2.1
An open-weight MiniMax model from December 2025 tuned for coding across many programming languages and app development
HY-Motion 1.0
Tencent's open model that turns text prompts into skeleton animations for 3D characters
Falcon H1R 7B
Small open reasoning model from Abu Dhabi research institute TII, aimed at maths, coding and science problems.
LFM2.5
Liquid AI's family of small open-weight models built to run on phones, laptops, cars and other devices
Step3-VL-10B
StepFun's small open-source model that understands images, charts and screens as well as text
GLM-Image
Open-source Z.ai image generator that is good at putting readable text into pictures
TranslateGemma
Google's open translation models built on Gemma 3, covering 55 languages in three sizes
SERA
Ai2's open coding agent models that can be trained cheaply on a private codebase and run in Claude Code
Z-Image
Alibaba Tongyi's open-weight 6-billion-parameter image model, the full base version behind Z-Image-Turbo.
Step-3.5-Flash
StepFun's open-source reasoning model for coding and agent tasks, cheap to run and free to download
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
Riverflow 2.0
Image generation and editing model built to keep product labels, text and fonts accurate
GPT-5.3-Codex
OpenAI coding model from February 2026 built for long agent tasks in Codex
Raven-1
Tavus's perception model that reads tone, facial expression and timing during live AI video calls
MiniMax M2.5
MiniMax's open-weight model for coding, tool use and office work, priced for long agent runs
FireRed-Image-Edit
Open-source image editing model from Xiaohongshu that follows written edit instructions while keeping faces and details consistent.
Tiny Aya
Cohere Labs' small open-weight multilingual model for translation and text in 70 languages, sized to run on phones
Mercury 2 (Inception)
Inception's diffusion-based reasoning model that writes over 1,000 tokens a second, aimed at voice agents and fast apps
QuiverAI Arrow
QuiverAI's family of models that generate, edit and animate SVG vector graphics such as logos
TADA
Hume AI's open-source text-to-speech model that keeps words and audio in step so it does not skip or invent words
GLM-5-Turbo
Closed Z.ai model tuned for OpenClaw agents, with tool use and long scheduled tasks
MiniMax M2.7
MiniMax's coding and agent model that helped run its own training experiments
MolmoWeb
An open Ai2 web agent model that uses a browser by looking at screenshots, then clicking, typing and scrolling
Cohere Transcribe
Open-source Cohere speech recognition model that turns audio in 14 languages into text
TRIBE v2
Meta research model that predicts how the human brain responds to images, sounds, video and text
Holo3
H Company's computer-use model that operates desktop software, with an open 35B version
LFM2.5-350M
Tiny 350-million-parameter Liquid AI model for tool calls and data extraction on cheap hardware
Trinity-Large-Thinking
Arcee AI's open-weight reasoning model built for long-running AI agents and multi-step tool use
Gemma 4
Google's open-weight model family in four sizes for reasoning and agents on local hardware
VOID
Netflix research model that removes an object from video along with the physical effects it caused
Wan2.7-Image
Alibaba's image model for making and editing pictures, with long text rendering in 12 languages and exact colour control.
GLM-5.1
Z.ai open-weights model built to run long coding tasks on its own for hours
Harrier
Microsoft's open-source text embedding models for search and retrieval in AI systems
HappyHorse
Alibaba's AI video model that makes clips with matching sound from text or images
ERNIE-Image
Open-weight Baidu text-to-image model with 8 billion parameters, strong at posters, comics and text in images.
Lyra 2.0
NVIDIA research model that turns a single image into a 3D world a camera can walk through
HY-World 2.0
Tencent's open world model that builds editable 3D scenes from text, images or video
Seed3D
ByteDance's model that turns a single image into a textured 3D asset ready for simulation
MiMo-V2.5-Pro
Xiaomi's trillion-parameter open-weight model for long coding and agent tasks with a one-million-token context
Echo-2
SpAItial model that turns a photo or text prompt into a 3D world that can be explored in real time
GPT-Realtime-2
OpenAI speech-to-speech API model for voice agents that can reason and use tools
ERNIE 5.1
Smaller, cheaper successor to ERNIE 5.0 from Baidu, tuned for agent tasks, reasoning and writing.
Command A+
Open-source Cohere language model for reasoning, images and agent tasks across 48 languages
Stable Audio 3.0
Stability AI's family of music and sound models, three of them free to download, trained on licensed data
MiniMax M3
MiniMax's open-weight model with a one-million-token context window, image and video input, and strong coding scores
MisoTTS (Miso One)
Miso Labs' open-weight 8B text-to-speech model that listens to the speaker's tone before it replies
North Mini Code
Cohere's first open-source coding model, a 30B mixture-of-experts model built for coding agents
DiffusionGemma
Experimental open Google model that writes blocks of text at once for faster local generation
GLM 5.2
Z.ai open-weights coding model with a 1M-token context, built for long agent coding runs
Sakana Fugu
Sakana AI's orchestration model that routes each task to a team of other AI models through one API
Leanstral 1.5
Mistral's free open model for writing and checking formal maths proofs in the Lean 4 language
Hy3
Tencent's open-weight language model for reasoning and agents, 295B parameters with 21B active
SWE-1.7
Cognition's earlier flagship coding model for Devin, built for long-running agent tasks at low cost
Ink-2
Cartesia's streaming speech-to-text model built for voice agents, with turn detection inside the model
Bonsai 27B
PrismML's heavily compressed 27B model small enough to run on a laptop or iPhone
Inkling
Thinking Machines Lab's first open-weight model, reading text, images and audio with adjustable thinking time
Lucy 2.5
Decart's video model that edits live video streams as they play, from text prompts or reference images
Seed Audio 1.0
ByteDance's audio model that makes speech, sound effects and ambience for a whole scene in one pass
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.
Laguna S 2.1
Poolside's open-weight coding model with a 1M-token context, small enough to run on local hardware
FLUX 3
Black Forest Labs model trained on images, video and audio together, generating video clips with sound up to 20 seconds.
Wan 3.0
Alibaba's video model that makes clips up to 30 seconds long from text, images, audio or office documents.
MAI-Cyber-1-Flash
Microsoft's compact security model for finding software vulnerabilities inside its MDASH code-scanning system
GLM-5.3
Z.ai flagship open-weights model with stronger coding and security research skills
FLUX Video Upscale
Black Forest Labs video upscaler that regenerates clips at up to 4K and repairs faces and textures.
Cohere Parse 5
Cohere vision model that turns PDFs, slides and scans into clean Markdown for search and AI tools
QuiverAI Arrow 2
QuiverAI's model for generating, vectorising and animating SVG graphics, released with a higher-end Telos version.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
SWE-2
Cognition's newest coding model for its Devin agents, with three reasoning-effort levels
Atria Dawn Preview
Open agentic model from Shanghai AI Laboratory for long research, coding and office tasks
Salesforce Koa
Salesforce's own reasoning model for Agentforce agents doing CRM work, built on NVIDIA Nemotron
StepAudio 3
StepFun's set of five audio models for voice agents, transcription, speech, sound and music
Holo4
H Company's agent models that use screens, code and API tools across desktop, web and phone
Mercury Voice (Inception)
Inception's diffusion model tuned for phone and voice agents, answering in about a third of a second
MAI-Transcribe-2-Streaming
Microsoft's real-time speech-to-text model for live captions and voice agents in 60 languages
EmbeddingGemma 2
Small open Google model that turns text, code, images, audio and video into searchable vectors on-device
Hunyuan-Vision-1.5-Thinking
Tencent's reasoning model for understanding images and video, offered through Tencent Cloud
MedGemma
Google's free open-weight models for reading medical images and text, built on Gemma 3 for health app developers
Chatterbox Turbo
Resemble AI's free, open-source English text-to-speech model with voice cloning and built-in watermarking
Nomos 1
Nous Research's open 30B model for solving maths problems and writing proofs in plain language
NousCoder-14B
A 14B open model from Nous Research trained with reinforcement learning for competitive programming
Jev
TypeSafe AI model that returns typed decisions with confidence scores for software automation, not text
GLM-OCR
Small open-source Z.ai model that pulls text, tables and formulas out of documents
MiMo-V2-Flash
Xiaomi's open-weight mixture-of-experts model built for fast reasoning, coding and agent work
Sonic 3
Cartesia's fast text-to-speech model for voice agents, with emotion, laughter and voice cloning