Every AI model release,
the moment it drops.

What ships next

Estimated from each model’s own release cadence. Overdue ones are the ones most likely to land this week.

• GPT Image

  OpenAI · now v5.4 · ships about every 147 days

  22 days overdue

• Kimi K2

  Moonshot AI · now v2.7 · ships about every 63 days

  54 days overdue

• GPT

  OpenAI · now v5.5 · ships about every 85 days

  82 days overdue

Predictions are estimates derived from public release history, not announcements.

Providers we track

• A

  Anthropic

  Claude — Opus, Sonnet and Haiku.

• O

  OpenAI

  GPT, o-series reasoning models, Sora and Whisper.

• G

  Google DeepMind

  Gemini, Gemma, Veo and Imagen.

• M

  Meta AI

  Llama — the open-weight family that started the wave.

• x

  xAI

  Grok, trained on Colossus.

• D

  DeepSeek

  Frontier reasoning at open-weight prices.

• A

  Alibaba Qwen

  Qwen — the widest open-weight lineup anywhere.

• M

  Mistral AI

  Mistral, Magistral, Devstral, Codestral.

• M

  Moonshot AI

  Kimi K-series, built for long context and agents.

• Z

  Z.AI

  GLM — Zhipu's open-weight coding and agent models.

• A

  Amazon

  Nova and Titan, native to Bedrock.

• M

  Microsoft

  Phi small models and the MAI in-house line.

• C

  Cohere

  Command and North, tuned for enterprise RAG.

• N

  NVIDIA

  Nemotron — open models tuned for their own silicon.

• M

  MiniMax

  MiniMax M-series, long context and video.

• B

  ByteDance

  Seed and Doubao, out of ByteDance's Seed lab.

• P

  Perplexity

  Sonar — search-grounded answers as an API.

• I

  IBM

  Granite — small, permissively licensed, enterprise-first.

• A

  AI21 Labs

  Jamba — hybrid Mamba/Transformer long context.

• A

  Allen Institute for AI

  OLMo and Molmo — fully open data, weights and code.

• R

  Reka AI

  Reka Core and Flash, natively multimodal.

• L

  Liquid AI

  Liquid Foundation Models — small enough for the edge.

• T

  Tencent

  Hunyuan — text, image and 3D generation.

• B

  Baidu

  ERNIE — Baidu's flagship multimodal line.

• S

  StepFun

  Step — multimodal models from Shanghai.

• U

  Upstage

  Solar — compact bilingual models from Korea.

• N

  Nous Research

  Hermes — community-tuned open models.

• I

  Inception Labs

  Mercury — diffusion language models, very fast.

• T

  Thinking Machines Lab

  Mira Murati's lab.

• S

  Sakana AI

  Nature-inspired model merging, out of Tokyo.

• X

  Xiaomi

  MiMo — on-device-first open models.

• A

  Arcee AI

  Small merged models for private deployment.

• I

  InclusionAI

  Ling and Ring, from Ant Group's open lab.

• S

  Sarvam AI

  Indic-language models built in India.

• P

  Poolside

  Frontier models trained specifically for software.

• W

  Writer

  Palmyra — domain-tuned enterprise models.

What you get

• Every lab, family and model version
• Follow a lab, a family, or one model
• Predicted next release dates
• Benchmarks, pricing and context side by side

Model releases are part of the AI Drops app, alongside the Claude Code changelog. For Claude Code slash commands, try /cards.