AI model lineage, source by source

The living family tree of AI models.

See who built each model, which family it belongs to, when it arrived, and what the official record actually says.

Open the Model Tree
Creators
44
Families
86
Releases
128
Sources
295
Featured ecosystems

Recorded lineage, creator by creator

Nesting shows a recorded predecessor Model Passport
CreatorAnthropicCreator of the Claude models, offered only as a hosted service through the Claude API and partner clouds.
Family

Claude 5

First released Jun 9, 2026

0 recorded lineage links in this family.

1 recorded lineage link leaves this family, named on the release it belongs to.

Creator positioning is recorded for 3 of the 4 variant names in Claude 5; the rest are marked as not recorded.

  • Opus tier — Anthropic states: For complex agentic coding and enterprise work

  • Sonnet tier — Anthropic states: The best combination of speed and intelligence

  • Fable tier — Anthropic states: Next-generation intelligence for long-running agents

    Continues into Claude Fable 5.1 (Claude 5.1), recorded in another family and shown without a connector.

  • Mythos tier — no creator statement of what this name is for is recorded.

Family

Claude 4.5

First released Sep 29, 2025

No recorded lineage links in this family, so no releases are connected here.

Creator positioning is recorded for the one variant name in Claude 4.5.

  • Haiku tier — Anthropic states: The fastest model with near-frontier intelligence

CreatorGoogle DeepMindThe Google research team that builds the Gemini models, formed by combining Google Brain and DeepMind.
Family

Gemini 3

First released Nov 18, 2025

1 recorded lineage link in this family.

Creator positioning is recorded for all 3 variant names in Gemini 3.

  • Flash tier — Google states: Gemini 3 Flash is our latest 3-series model, with Pro-level intelligence at the speed and pricing of Flash.

  • Flash tier — Google states: Gemini 3 Flash is our latest 3-series model, with Pro-level intelligence at the speed and pricing of Flash.

  • Flash tier — Google states: Gemini 3 Flash is our latest 3-series model, with Pro-level intelligence at the speed and pricing of Flash.

  • Flash-Lite tier — Google states: a low-latency, cost-effective multimodal model optimized for high-frequency, lightweight tasks

    • Flash-Lite tier — Google states: a low-latency, cost-effective multimodal model optimized for high-frequency, lightweight tasks

  • Pro tier — Google states: best for complex tasks that require broad world knowledge and advanced reasoning across modalities

Family

Gemini 2.5

First released Jun 17, 2025

0 recorded lineage links in this family.

Creator positioning is recorded for all 2 variant names in Gemini 2.5.

  • Flash tier — Google states: Our best price-performance model for low-latency, high-volume tasks that require reasoning.

  • Pro tier — Google states: Our most advanced model for complex tasks, featuring deep reasoning and coding capabilities in the 2.5 family.

CreatorMetaCreator of two model lines that reach people differently, so not every Meta model is downloadable. Llama is published as downloadable weights under Meta's own community licences, collected at huggingface.co/meta-llama. Muse, developed by Meta Superintelligence Labs, is served only through Meta's own surfaces — the Meta AI app, meta.ai, and the Meta Model API — and no Muse weights are released. Both lines are announced on AI at Meta, which is what website and releasePage record: it is an announcement stream rather than an index of releases, and nothing checked here shows a single Meta page that lists both lines. The Hugging Face collection is Llama-only and is recorded as a source rather than as this creator's release page. The former Llama-only site www.llama.com now redirects to developer.meta.com and is no longer recorded here.
Family

Muse

First released Apr 8, 2026

1 recorded lineage link in this family.

No creator statement of what these 3 variant names mean is recorded for Muse.

Family

Llama 4

First released Apr 5, 2025

0 recorded lineage links in this family.

No creator statement of what these 2 variant names mean is recorded for Llama 4.

Family

Llama 3

First released Jul 23, 2024

No recorded lineage links in this family, so no releases are connected here.

No creator statement of what these 6 variant names mean is recorded for Llama 3.

CreatorMicrosoftA company that occupies three roles this dataset keeps apart. It creates model families, through Microsoft AI, which builds the MAI line, and through Microsoft Research, which publishes the Fara line; those are the only Microsoft records treated as models. It ships Microsoft Copilot, which is a product and is recorded as one. It operates Microsoft Foundry, which is a serving platform and is recorded as one. A model appearing in Microsoft Foundry is a deployment fact and never evidence that Microsoft created it.
Family

MAI

First released Aug 28, 2025

No recorded lineage links in this family, so no releases are connected here.

No creator statement of what these 5 variant names mean is recorded for MAI.

CreatorOpenAICreator of the GPT-4.1 API family and the GPT-5 line (GPT-5, GPT-5.4, GPT-5.5, and the current GPT-5.6 family of Sol, Terra, and Luna).
Family

GPT-5.6

First released Jul 9, 2026

0 recorded lineage links in this family.

No creator statement of what these 4 variant names mean is recorded for GPT-5.6.

Family

GPT-4.1

First released Apr 14, 2025

0 recorded lineage links in this family.

No creator statement of what these 3 variant names mean is recorded for GPT-4.1.

Families without a featured release are not drawn here. Every record in the catalog, featured or not, is listed in the model index and the Model Tree.

Release pulse

Recent verified changes

Source-dated lifecycle changes from the last 18 months, newest first, at most 6. Dated by what the record says happened, not by when we last re-checked it.

  1. Generally available

    Anthropic launched Claude Fable 5.1 and described it as generally available, on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry.

  2. API available

    Anthropic launched Claude Mythos 5.1 on the Claude API alongside Claude Fable 5.1, for Project Glasswing participants. Anthropic describes it as available only through its trusted access programs rather than generally available.

  3. API available

    Amazon announced Alibaba Cloud’s Qwen3.8-27B as available on Amazon SageMaker JumpStart. This changes where the model can be deployed and changes nothing about who created it.

  4. Generally available

    Alibaba Cloud recorded Qwen3.8-27B as available on Hugging Face Hub and ModelScope.

  5. Preview

    Microsoft AI announced MAI-Thinking-1 as available in public preview on Microsoft Foundry.

  6. Generally available

    Alibaba Cloud recorded Qwen3.8-2.4T-A95B as available on Hugging Face Hub and ModelScope.

Tracking 128 releases across 86 families from 44 creators, each backed by primary sources (295 in all). A curated seed, not a complete census of the field.