How the major AI models are actually related

Every edge in the Claude tree above, in chronological order. Each line shows: destination model, edge type, source model, and the provider documentation that establishes the relationship (where one was disclosed).

• Claude 2 (Jul 11, 2023) — succession of Claude 1
• Claude Instant 1.2 (Aug 9, 2023) — distilled from of Claude 1 — Anthropic positioned Claude Instant as a smaller, faster sibling of the Claude flagship; sourced to the Claude 2 announcement.
• Claude 2.1 (Nov 21, 2023) — post-training of Claude 2 — Anthropic documented Claude 2.1 as an iterative update on Claude 2 with the 200K context extension, not a new base.
• Claude 3 Opus (Mar 4, 2024) — new base of Claude 2.1 — Anthropic introduced the Claude 3 family as a new generation of base models, not a Claude 2 fine-tune.
• Claude 3 Sonnet (Mar 4, 2024) — succession of Claude 3 Opus — Anthropic shipped Opus / Sonnet / Haiku as the three tiers of the Claude 3 generation simultaneously.
• Claude 3 Haiku (Mar 13, 2024) — succession of Claude 3 Opus
• Claude 3.5 Sonnet (Jun 20, 2024) — succession of Claude 3 Sonnet
• 3.5 Sonnet (new) (Oct 22, 2024) — post-training of Claude 3.5 Sonnet — Anthropic documented the October 2024 'new 3.5 Sonnet' as an upgraded version under the same model id.
• Claude 3.5 Haiku (Nov 4, 2024) — distilled from of 3.5 Sonnet (new) — Anthropic documented Claude 3.5 Haiku as a smaller sibling sharing the 3.5 generation training.
• Claude 3.7 Sonnet (Feb 24, 2025) — succession of 3.5 Sonnet (new)
• Claude Sonnet 4 (May 22, 2025) — new base of Claude 3.7 Sonnet — Anthropic introduced Claude 4 as a new generation, with Opus 4 and Sonnet 4 as new bases rather than fine-tunes of 3.7.
• Claude Opus 4 (May 22, 2025) — new base of Claude 3.7 Sonnet
• Claude Opus 4.1 (Aug 5, 2025) — succession of Claude Opus 4
• Claude Sonnet 4.5 (Sep 29, 2025) — succession of Claude Sonnet 4
• Claude Haiku 4.5 (Oct 15, 2025) — succession of Claude 3.5 Haiku
• Claude Opus 4.5 (Nov 24, 2025) — succession of Claude Opus 4.1
• Claude Opus 4.6 (Feb 5, 2026) — succession of Claude Opus 4.5
• Claude Sonnet 4.6 (Feb 17, 2026) — succession of Claude Sonnet 4.5
• Claude Opus 4.7 (Apr 16, 2026) — succession of Claude Opus 4.6
• Claude Opus 4.8 (May 28, 2026) — succession of Claude Opus 4.7
• Claude Fable 5 (Jun 9, 2026) — succession of Claude Opus 4.8 — Anthropic introduced Claude Fable 5 (June 9, 2026) as the first generally-available model in the new Mythos class — a tier positioned above Opus in capability. Claude Mythos 5 (claude-mythos-5) is disclosed as the same underlying model as Fable 5 with safeguards lifted in some areas — restricted to Project Glasswing cyberdefenders and trusted-access partners, not a public GA model, so it is not drawn as a separate node. Opus 4.8 is described as the 'next-most-capable model' that Fable 5 falls back to on safeguard-triggered queries; Anthropic does not disclose Fable 5 as a fine-tune, post-training, or distillation of Opus 4.8. Suspended June 12 under a US export-control directive and redeployed globally July 1, 2026 — per the anthropic.com/news/claude-fable-5-mythos-5 announcement and the July 1 redeployment update.
• Claude Sonnet 5 (Jun 30, 2026) — succession of Claude Sonnet 4.6
• Claude Opus 5 (Jul 24, 2026) — succession of Claude Opus 4.8 — Claude Opus 5 (July 24, 2026) is the new head of the Opus tier, described by Anthropic as coming 'close to the frontier intelligence of Claude Fable 5 at half the price' and as providing 'greatly improved performance for the same cost as its predecessor, Opus 4.8.' It is the new default model on Claude Max and the strongest model on Claude Pro, and it remains behind Mythos 5 on cybersecurity tasks. Anthropic does not disclose Opus 5 as a fine-tune, post-training, or distillation of Opus 4.8 — per the anthropic.com/news/claude-opus-5 announcement.

Every edge in the ChatGPT tree above, in chronological order. Each line shows: destination model, edge type, source model, and the provider documentation that establishes the relationship (where one was disclosed).

• GPT-3.5 (Nov 30, 2022) — post-training of GPT-3 — OpenAI's GPT-3.5 (text-davinci series and ChatGPT launch) was an RLHF-fine-tuned descendant of the GPT-3 base, per OpenAI's InstructGPT paper and the ChatGPT launch post.
• GPT-4 (Mar 14, 2023) — new base of GPT-3.5 — OpenAI introduced GPT-4 as a new base model, not a GPT-3.5 fine-tune.
• GPT-4 Turbo (Nov 6, 2023) — post-training of GPT-4 — OpenAI documented GPT-4 Turbo at DevDay 2023 as the same generation with extended context and updated training data.
• GPT-4o (May 13, 2024) — new base of GPT-4 Turbo — OpenAI introduced GPT-4o as a new natively-multimodal base model.
• GPT-4o mini (Jul 18, 2024) — distilled from of GPT-4o — OpenAI positioned GPT-4o mini as the smaller, distilled sibling of GPT-4o.
• o1-preview (Sep 12, 2024) — new base of GPT-4o — OpenAI introduced the o-series as a separate reasoning track; o1 was trained to use long chain-of-thought at inference time.
• o1 (Dec 5, 2024) — succession of o1-preview
• o3-mini (Jan 31, 2025) — succession of o1
• GPT-4.1 (Apr 14, 2025) — succession of GPT-4o
• o4-mini (Apr 16, 2025) — distilled from of o3 — OpenAI positioned o4-mini as the smaller sibling shipped alongside o3.
• o3 (Apr 16, 2025) — succession of o3-mini
• GPT-5 (Aug 7, 2025) — lines merged of GPT-4.1 — OpenAI introduced GPT-5 as a unified router that combines the GPT chat line and the o-series reasoning track into one model id.
• GPT-5 (Aug 7, 2025) — lines merged of o3
• GPT-5.1 (Nov 12, 2025) — succession of GPT-5
• GPT-5.2 (Dec 11, 2025) — succession of GPT-5.1
• GPT-5.3-Codex (Feb 5, 2026) — succession of GPT-5.2
• GPT-5.3 Instant (Mar 3, 2026) — post-training of GPT-5.2 — OpenAI documented GPT-5.3 Instant (March 3, 2026) as a ChatGPT default update focused on conversational quality, separate from the Codex-specialized GPT-5.3-Codex shipped two days earlier — sourced to the GPT-5.3 Instant announcement at openai.com/index/gpt-5-3-instant/.
• GPT-5.4 (Mar 5, 2026) — succession of GPT-5.3-Codex
• GPT-5.5 (Apr 23, 2026) — succession of GPT-5.4
• GPT-5.6 Sol (Jul 9, 2026) — succession of GPT-5.5 — GPT-5.6 (Sol / Terra / Luna) previewed June 26, 2026 and reached general availability July 9, 2026 across ChatGPT, Codex, and the OpenAI API, replacing GPT-5.5 as OpenAI's current frontier flagship. It introduces a new naming system: the number denotes the generation, while Sol (flagship), Terra (balanced), and Luna (fast, low-cost) are durable capability tiers. OpenAI's model catalog at developers.openai.com/api/docs/models lists gpt-5.6-sol as the 'frontier model for complex professional work' (aliased gpt-5.6, 1.05M context, February 16, 2026 knowledge cutoff) alongside gpt-5.6-terra and gpt-5.6-luna. OpenAI does not disclose GPT-5.6 as a fine-tune, post-training, or distillation of GPT-5.5 — per the openai.com/index/gpt-5-6/ GA announcement (preview: openai.com/index/previewing-gpt-5-6-sol/).

Every edge in the Gemini tree above, in chronological order. Each line shows: destination model, edge type, source model, and the provider documentation that establishes the relationship (where one was disclosed).

• PaLM 2 (May 10, 2023) — new base of Bard (LaMDA) — Google introduced PaLM 2 as a new base model powering an upgraded Bard.
• Gemini 1.0 Pro (Dec 6, 2023) — new base of PaLM 2 — Google's Gemini 1.0 was a new natively-multimodal architecture, not a PaLM 2 fine-tune.
• Gemini 1.0 Ultra (Feb 8, 2024) — succession of Gemini 1.0 Pro
• Gemini 1.5 Pro (Feb 15, 2024) — new base of Gemini 1.0 Ultra — Google introduced Gemini 1.5 with a new mixture-of-experts architecture and the 1M context, not a 1.0 fine-tune.
• Gemini 1.5 Flash (May 14, 2024) — distilled from of Gemini 1.5 Pro — Google described Gemini 1.5 Flash as a distilled smaller sibling of 1.5 Pro.
• Gemini 2.0 Flash (Dec 11, 2024) — succession of Gemini 1.5 Flash
• Gemini 2.0 Pro Exp (Feb 5, 2025) — succession of Gemini 1.5 Pro
• Gemini 2.5 Pro (Mar 25, 2025) — succession of Gemini 2.0 Pro Exp
• Gemini 2.5 Flash (Jun 17, 2025) — succession of Gemini 2.0 Flash
• 2.5 Flash-Lite (Jul 22, 2025) — distilled from of Gemini 2.5 Flash — Google positioned Flash-Lite as the smallest sibling of the 2.5 Flash family.
• Gemini 3 Pro (Nov 18, 2025) — new base of Gemini 2.5 Pro — Google introduced Gemini 3 as a new generation flagship.
• Gemini 3 Flash (Dec 17, 2025) — succession of Gemini 2.5 Flash
• Gemini 3.1 Pro (Feb 19, 2026) — succession of Gemini 3 Pro
• 3.1 Flash-Lite (Mar 3, 2026) — succession of Gemini 3 Flash — Google positions Flash-Lite as the fastest, most cost-effective member of each Flash generation rather than as a distilled student of the Flash model above it; the July 21, 2026 Flash announcement compares 3.5 Flash-Lite against 'prior Flash-Lite generations' and against 3 Flash on benchmarks, and discloses no teacher–student relationship. Softened from 'distilled from' on the 2026-07-31 sweep because no Google source substantiates the stronger claim.
• Gemini 3.5 Flash (May 19, 2026) — new base of Gemini 3 Flash — Google introduced Gemini 3.5 Flash (May 19, 2026) at Google I/O as the first Flash to outperform the prior generation's Pro flagship on hard coding and agentic benchmarks — a new generation in the Flash track. Per the blog.google announcement, it powers Gemini Spark and is the new default behind the Gemini app and Search AI Mode.
• Gemini 3.6 Flash (Jul 21, 2026) — succession of Gemini 3.5 Flash — Gemini 3.6 Flash reached Stable GA July 21, 2026 as Google's workhorse Flash model, replacing 3.5 Flash as the current Gemini frontier. Google says it 'builds directly on developer and customer feedback from 3.5 Flash' — consuming 17% fewer output tokens on the Artificial Analysis Index at a lower price ($1.50 / $7.50 per million input / output tokens) while improving on DeepSWE (49% vs. 37%), MLE-Bench (63.9% vs. 49.7%) and OSWorld-Verified (83.0% vs. 78.4%). That is a feedback-and-benchmarks framing, not a disclosed base-model relationship, so the edge stays a succession — per blog.google 'Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber' (July 21, 2026).
• 3.5 Flash-Lite (Jul 21, 2026) — succession of Gemini 3.5 Flash — Gemini 3.5 Flash-Lite shipped the same day as 3.6 Flash (July 21, 2026) as 'our fastest, most cost-effective 3.5-class model' at 350 output tokens per second and $0.30 / $2.50 per million tokens. Google frames it as the low-latency, high-throughput member of the 3.5 series and benchmarks it against 3.1 Flash-Lite and 3 Flash, but does not describe it as distilled from 3.5 Flash — per the same July 21, 2026 blog.google announcement.

Every edge in the Grok tree above, in chronological order. Each line shows: destination model, edge type, source model, and the provider documentation that establishes the relationship (where one was disclosed).

• Grok 1.5 (Mar 28, 2024) — post-training of Grok 1 — xAI documented Grok 1.5 as a successor that extended Grok 1's context and reasoning, broadly within the same generation.
• Grok 2 (Aug 13, 2024) — new base of Grok 1.5 — xAI introduced Grok 2 as a new base.
• Grok 3 (Feb 17, 2025) — new base of Grok 2 — xAI introduced Grok 3 trained on the Memphis Colossus cluster as a new base.
• Grok 4 (Jul 9, 2025) — new base of Grok 3
• grok-code-fast-1 (Aug 28, 2025) — new base of Grok 4 — xAI documented grok-code-fast-1 (August 28, 2025) as a from-scratch architecture optimized for agentic coding rather than a fine-tune of Grok 4 — per the x.ai/news/grok-code-fast-1 announcement.
• Grok 4 Fast (Sep 19, 2025) — distilled from of Grok 4 — xAI positioned Grok 4 Fast (September 19, 2025) as a faster, cheaper Grok 4 with a 2M-token context window — per the x.ai/news/grok-4-fast announcement.
• Grok 4.1 (Nov 17, 2025) — post-training of Grok 4
• Grok 4.1 Fast (Nov 19, 2025) — distilled from of Grok 4.1 — xAI positioned Grok 4.1 Fast as a smaller, faster sibling shipped two days after 4.1.
• Grok 4.20 (Mar 10, 2026) — succession of Grok 4.1
• Grok 4.3 (Apr 17, 2026) — succession of Grok 4.20
• Grok Build 0.1 (May 19, 2026) — succession of grok-code-fast-1 — xAI documented Grok Build 0.1 (early-access launch May 19, 2026; the Grok Build CLI / TUI launched in beta five days earlier on May 14, 2026) as the successor to grok-code-fast-1, purpose-built for agentic coding workflows — per the x.ai/news/grok-build-0-1 announcement and the May 15, 2026 deprecation of grok-code-fast-1.
• Composer 2.5 (Jun 1, 2026) — succession of Grok Build 0.1 — xAI shipped Composer 2.5 (June 1, 2026) inside the Grok Build /model menu as a fast agentic-coding sibling to Grok Build 0.1; the original launch coverage identified Composer 2.5 as built on the open-source Kimi K2.5 checkpoint (Moonshot AI) and post-trained with roughly 25× more synthetic agentic tasks than Composer 2, so the line of descent is external — xAI does not disclose Composer 2.5 as a fine-tune, post-training, or distillation of any prior xAI model. The x.ai/news/composer-2-5 page itself has since been pared back; the Kimi K2.5 attribution is preserved on /ai/grok/versions/#grok-composer-2-5 with the original launch citation.
• Grok 4.5 (Jul 8, 2026) — succession of Grok 4.3 — Grok 4.5 (July 8, 2026) is xAI's newest chat and coding flagship, collapsing the Grok 4.x line's separate reasoning and non-reasoning SKUs into a single grok-4.5 model id with a low / medium / high reasoning-effort setting. xAI describes it as its 'strongest model ever,' trained across tens of thousands of NVIDIA GB300 GPUs on new coding / science / engineering / math data; it does not disclose Grok 4.5 as a fine-tune, post-training, or distillation of Grok 4.3 — per the x.ai/news/grok-4-5 announcement.

Every edge in the Llama tree above, in chronological order. Each line shows: destination model, edge type, source model, and the provider documentation that establishes the relationship (where one was disclosed).

• Llama 2 (Jul 18, 2023) — new base of LLaMA 1 — Meta's Llama 2 paper documents it as a new pretraining run with substantially larger data and the open license shift.
• Code Llama (Aug 24, 2023) — fine-tune of Llama 2 — Meta's Code Llama paper documents it as a fine-tune of Llama 2 on code data.
• Code Llama 70B (Jan 29, 2024) — fine-tune of Code Llama — Meta's Code Llama 70B release (January 29, 2024) is described as a continued-pretrained / fine-tuned 70B variant of the Code Llama line, sourced from the Llama 2 70B base — per the ai.meta.com Code Llama 70B announcement.
• Llama 3 (Apr 18, 2024) — new base of Llama 2 — Meta's Llama 3 release notes document it as a new pretraining run with updated tokenizer.
• Llama 3.1 (Jul 23, 2024) — post-training of Llama 3 — Meta's Llama 3.1 release added the 405B size and the long-context post-training; documented as same generation.
• Llama Guard 3 (Jul 23, 2024) — fine-tune of Llama 3.1 — Meta's Llama Guard 3 family (July 23, 2024, alongside Llama 3.1) is a safeguard fine-tune of the Llama 3.1 base model line — per the Llama 3.1 release notes documenting Guard 3 8B / 1B / 11B-Vision as classifier fine-tunes.
• Llama 3.2 (Sep 25, 2024) — post-training of Llama 3.1 — Meta's Llama 3.2 release added vision and edge-sized variants.
• Llama 3.3 70B (Dec 6, 2024) — post-training of Llama 3.2
• Llama 4 Scout (Apr 5, 2025) — new base of Llama 3.3 70B — Meta's Llama 4 release introduced a new mixture-of-experts architecture as a new base.
• Llama 4 Maverick (Apr 5, 2025) — new base of Llama 3.3 70B
• Llama Guard 4 (Apr 29, 2025) — fine-tune of Llama 4 Maverick — Meta's Llama Guard 4 12B (April 29, 2025) is a multimodal safeguard fine-tune released alongside the Llama 4 family — per the Llama 4 family release notes documenting Guard 4 as a Llama-4-line classifier.
• Muse Spark (Apr 8, 2026) — new base of Llama 4 Maverick — Meta described Muse Spark as the closed-weights successor to the Llama line, trained at Meta Superintelligence Labs.
• Muse Spark 1.1 (Jul 9, 2026) — succession of Muse Spark — Muse Spark 1.1 (July 9, 2026) is Meta Superintelligence Labs' next Muse Spark release — a multimodal reasoning model for agentic tasks with a 1M actively-managed context, launched consumer-GA in the Meta AI app ('Thinking' mode) plus a paid Meta Model API in public preview. Meta describes it as 'a significant upgrade from Muse Spark' but does not disclose it as a fine-tune, post-training, or distillation of Muse Spark — per the ai.meta.com/blog/introducing-muse-spark-meta-model-api/ announcement.
• Muse Spark 1.2 (Aug 5, 2026) — succession of Muse Spark 1.1 — Muse Spark 1.2 (August 5, 2026) is described by Meta Superintelligence Labs as 'a coding-focused update to Muse Spark 1.1', co-trained with the Muse Code terminal agent and trained through a self-improvement loop in which Muse Spark 1.1 generated and graded the coding environments used as training data. Meta does not disclose it as a fine-tune, post-training, or distillation of the 1.1 checkpoint, so the edge stays succession — per the research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2 announcement.

Every edge in the DeepSeek tree above, in chronological order. Each line shows: destination model, edge type, source model, and the provider documentation that establishes the relationship (where one was disclosed).

• DeepSeek-Coder (Nov 2, 2023) — fine-tune of DeepSeek-LLM — DeepSeek-Coder shipped alongside the base LLM as a code-specialized fine-tune.
• DeepSeek-V2 (May 6, 2024) — new base of DeepSeek-LLM — DeepSeek-V2 paper introduced the DeepSeekMoE architecture and Multi-head Latent Attention as new base.
• DeepSeek-V3 (Dec 26, 2024) — new base of DeepSeek-V2 — DeepSeek-V3 paper introduced the 671B-parameter MoE as a new base.
• DeepSeek-R1 (Jan 20, 2025) — post-training of DeepSeek-V3 — DeepSeek-R1 paper documents R1 as RL post-training over the V3 base — RL-only emergent chain-of-thought.
• DeepSeek-V3-0324 (Mar 24, 2025) — post-training of DeepSeek-V3 — DeepSeek-V3-0324 release (March 24, 2025) is documented as an MIT-relicensed update to the V3 base with improved reasoning / coding / tool use — per the HuggingFace card at deepseek-ai/DeepSeek-V3-0324.
• DeepSeek-R1-0528 (May 28, 2025) — post-training of DeepSeek-R1 — DeepSeek-R1-0528 (May 28, 2025) is documented as an R1 update with the same 671B / 37B-active MoE architecture, improved reasoning depth and tool use — per the HuggingFace card at deepseek-ai/DeepSeek-R1-0528.
• DeepSeek-V3.1 (Aug 21, 2025) — lines merged of DeepSeek-R1-0528 — DeepSeek documented V3.1's hybrid Thinking / Non-Thinking architecture as the architectural convergence of the V-series and the standalone R-series — no further standalone R-series releases shipped after R1-0528.
• DeepSeek-V3.1 (Aug 21, 2025) — post-training of DeepSeek-V3-0324 — DeepSeek-V3.1 (August 21, 2025) merged the V-series and R-series into a hybrid Thinking / Non-Thinking architecture — per the api-docs.deepseek.com/news/news250821 release note (the generic api-docs.deepseek.com thinking-mode guide was subsequently rewritten around V4-pro).
• DeepSeek-V3.2-Exp (Sep 29, 2025) — new base of DeepSeek-V3.1 — DeepSeek-V3.2-Exp (September 29, 2025) is described by DeepSeek as 'an experimental version of our model' and 'an intermediate step toward our next-generation architecture' that 'builds upon V3.1-Terminus by introducing DeepSeek Sparse Attention' — per the github.com/deepseek-ai/DeepSeek-V3.2-Exp README.
• DeepSeek-V3.2 (Dec 1, 2025) — post-training of DeepSeek-V3.2-Exp — DeepSeek-V3.2 stable (December 1, 2025) productionized the DSA recipe from V3.2-Exp into the 685B-parameter MoE flagship — per the arXiv 2512.02556 technical paper.
• DeepSeek-V4-Pro (Apr 24, 2026) — new base of DeepSeek-V3.2 — DeepSeek-V4 introduced a new MoE base with Thinking and Non-Thinking modes.
• DeepSeek-V4-Flash (Apr 24, 2026) — distilled from of DeepSeek-V4-Pro — DeepSeek positioned V4-Flash as the smaller sibling shipped alongside V4-Pro.

Every edge in the Mistral tree above, in chronological order. Each line shows: destination model, edge type, source model, and the provider documentation that establishes the relationship (where one was disclosed).

• Mixtral 8x7B (Dec 11, 2023) — new base of Mistral 7B — Mixtral 8x7B introduced the sparse mixture-of-experts architecture as a new base, not a Mistral 7B fine-tune.
• Mistral Large (Feb 26, 2024) — new base of Mixtral 8x7B — Mistral Large was released as a proprietary flagship under the Mistral Research / commercial license, not a Mixtral fine-tune.
• Mistral Large 2 (Jul 24, 2024) — new base of Mistral Large
• Codestral 25.01 (Jan 13, 2025) — succession of Mistral Large 2 — Codestral 25.01 (January 13, 2025) is documented as the next-generation Codestral coding flagship — per mistral.ai/news/codestral-2501. Mistral does not disclose Codestral 25.01 as a fine-tune, post-training, or distillation of Mistral Large 2 (the Codestral line is its own track since the original Codestral 22B, May 2024).
• Mistral Small 3 (Jan 30, 2025) — new base of Mistral Large 2 — Mistral Small 3 marked the December 2025 'Mistral 3' family relaunch under Apache 2.0 with new bases.
• Mistral Medium 3 (May 7, 2025) — succession of Mistral Small 3 — Mistral Medium 3 (May 7, 2025) is the mid-tier proprietary flagship — per mistral.ai/news/mistral-medium-3.
• Devstral Small (May 21, 2025) — fine-tune of Mistral Small 3 — Devstral Small (May 21, 2025) is a 24B coding-agent fine-tune of Mistral Small 3.1: the model card declares base_model mistralai/Mistral-Small-3.1-24B-Instruct-2503 and states 'It is finetuned from Mistral-Small-3.1 … before fine-tuning from Mistral-Small-3.1 the vision encoder was removed' — per the HuggingFace card at mistralai/Devstral-Small-2505. The mistral.ai/news/devstral launch post itself now names Mistral Small 3.1 only in a pricing comparison, so the model card is the surviving disclosure.
• Magistral (Jun 10, 2025) — fine-tune of Mistral Small 3 — Magistral Small (June 10, 2025) declares base_model mistralai/Mistral-Small-3.1-24B-Instruct-2503 and is described as 'Building upon Mistral Small 3.1 (2503), with added reasoning capabilities, undergoing SFT from Magistral Medium traces and RL on top' — per the HuggingFace card at mistralai/Magistral-Small-2506. The launch post adds that 'Magistral is fine-tuned for multi-step logic' (mistral.ai/news/magistral; arXiv 2506.10910).
• Codestral 25.08 (Jul 30, 2025) — succession of Codestral 25.01 — Codestral 25.08 (July 30, 2025) is the next Codestral flagship, announced with +30% accepted completions and 50% fewer runaway generations over prior versions — per mistral.ai/news/codestral-25-08.
• Mistral Large 3 (Dec 2, 2025) — succession of Mistral Medium 3
• Ministral 3 (Dec 2, 2025) — succession of Mistral Small 3 — Ministral 3 (December 2, 2025) is the small-end open-weights line in the Mistral 3 family relaunch — per the mistral.ai/news/mistral-3 announcement.
• Devstral 2 (Dec 9, 2025) — succession of Devstral Small — Devstral 2 (December 9, 2025) is 'our next-generation coding model family available in two sizes: Devstral 2 (123B) and Devstral Small 2 (24B)' — per mistral.ai/news/devstral-2-vibe-cli.
• Mistral OCR 3 (Dec 17, 2025) — succession of Mistral Large 3 — Mistral OCR 3 (December 17, 2025) ships alongside the Mistral 3 family on la Plateforme as the upgraded structured-document model; Mistral's announcement positions OCR 3 as a major upgrade over Mistral OCR 2 (74% win rate) and does not document any lineage from Mistral Large 3 — per mistral.ai/news/mistral-ocr-3. The OCR-3 ↔ Mistral 3 family edge here is a co-shipped successor relationship, not a disclosed base-model claim.
• Mistral Small 4 (Mar 16, 2026) — succession of Mistral Large 3
• Leanstral (Mar 16, 2026) — succession of Mistral Small 4 — Leanstral (March 16, 2026) is 'the first open-source code agent designed for Lean 4' and was 'built as part of the Mistral Small 4 family' — a 119B / 6.5B-active MoE under Apache 2.0. Mistral describes family membership rather than a fine-tune or distillation of Mistral Small 4 — per the HuggingFace card at mistralai/Leanstral-2603 and mistral.ai/news/leanstral.
• Mistral Med 3.5 (Apr 28, 2026) — new base of Mistral Small 4 — Mistral Medium 3.5 (April 28, 2026) is 'our first flagship merged model … a dense 128B model with a 256k context window, handling instruction-following, reasoning, and coding in a single set of weights,' with a vision encoder Mistral says it 'trained from scratch.' The card names Mistral Medium 3.1 and Magistral as the models it replaces in Le Chat and Devstral 2 in Vibe — per the HuggingFace card at mistralai/Mistral-Medium-3.5-128B and the model card at docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04.
• Mistral OCR 4 (Jun 23, 2026) — succession of Mistral OCR 3 — Mistral OCR 4 (June 23, 2026) is the next document-intelligence flagship, adding paragraph-level bounding boxes, typed block classification and inline confidence scores over the text-extraction focus of prior generations, and is benchmarked head-to-head against 'our own Mistral OCR 3.' Mistral discloses no base-model relationship between the two — per mistral.ai/news/ocr-4.
• Leanstral 1.5 (Jun 30, 2026) — post-training of Leanstral — Leanstral 1.5 (June 30, 2026) declares base_model mistralai/Leanstral-2603 and is introduced as 'an updated model of our previously released Leanstral model,' trained through mid-training, supervised fine-tuning, and reinforcement learning with CISPO on the same 119B / 6.5B-active architecture — per the HuggingFace card at mistralai/Leanstral-1.5-119B-A6B and mistral.ai/news/leanstral-1-5.

Every edge in the Qwen tree above, in chronological order. Each line shows: destination model, edge type, source model, and the provider documentation that establishes the relationship (where one was disclosed).

• Qwen-14B (Sep 25, 2023) — succession of Qwen-7B
• Qwen2 family (Jun 7, 2024) — new base of Qwen-14B — Qwen2 marked the June 2024 architecture refresh with the Apache 2.0 licensing turn for most variants.
• Qwen2.5 family (Sep 19, 2024) — post-training of Qwen2 family
• QwQ-32B Preview (Nov 28, 2024) — fine-tune of Qwen2.5 family — QwQ-32B-Preview (November 28, 2024) is the first Qwen reasoning model — 'an experimental research model developed by the Qwen Team, focused on advancing AI reasoning capabilities,' a 32B dense Apache 2.0 release whose model card declares base_model Qwen/Qwen2.5-32B-Instruct. Per the HuggingFace card at Qwen/QwQ-32B-Preview and the launch post at qwenlm.github.io/blog/qwq-32b-preview.
• Qwen2.5-Max (Jan 29, 2025) — succession of Qwen2.5 family — Qwen2.5-Max (January 29, 2025) is the first proprietary closed-weights Qwen-Max release — per qwenlm.github.io/blog/qwen2.5-max.
• QwQ-32B (Mar 5, 2025) — post-training of QwQ-32B Preview — QwQ-32B (March 5, 2025) is the production successor to QwQ-32B-Preview, trained with RL for chain-of-thought reasoning — per qwenlm.github.io/blog/qwq-32b.
• Qwen3 family (Apr 28, 2025) — new base of Qwen2.5 family — Qwen3 introduced the hybrid Thinking / Non-Thinking architecture, absorbing the standalone QwQ reasoning track.
• Qwen3 family (Apr 28, 2025) — lines merged of QwQ-32B — Qwen3 absorbed the QwQ standalone-reasoning track into the hybrid Thinking / Non-Thinking architecture — no further standalone QwQ rows are expected per qwen3 release post.
• Qwen3-Coder (Jul 22, 2025) — succession of Qwen3 family — Qwen3-Coder (July 22, 2025) is Alibaba's open-weights agentic-coding flagship, led by Qwen3-Coder-480B-A35B-Instruct. The launch post documents its own pre-training run — 7.5T tokens at a 70% code ratio, native 256K context extended to 1M with YaRN, and scaled code RL in post-training — rather than a fine-tune of a Qwen3 chat checkpoint, so the edge stays a succession within the Qwen3 generation. Softened from 'fine-tune' on the 2026-07-31 sweep; per qwenlm.github.io/blog/qwen3-coder.
• Qwen3-Next 80B (Sep 11, 2025) — new base of Qwen3 family — Qwen3-Next-80B-A3B (September 11, 2025) introduced a novel ultra-sparse MoE architecture (hybrid Gated DeltaNet + Gated Attention) as a new base — per the Alibaba Cloud Community blog post.
• Qwen3-Max (Sep 15, 2025) — succession of Qwen3 family — Qwen3-Max is the largest Qwen3-generation proprietary flagship.
• Qwen3-VL (Sep 23, 2025) — fine-tune of Qwen3 family — Qwen3-VL family (September 23, 2025) is the vision-language fine-tune of Qwen3, mirroring the hybrid-mode recipe at the vision-language layer — per arXiv 2511.21631.
• Qwen3-Coder-Next (Feb 4, 2026) — fine-tune of Qwen3-Coder — Qwen3-Coder-Next (February 4, 2026) is built on Qwen3-Next-80B-A3B-Base, the ultra-sparse coding-agent line — per qwen.ai/blog?id=qwen3-coder-next.
• Qwen3.5 + Plus (Feb 16, 2026) — succession of Qwen3-Next 80B — Qwen3.5 (February 16, 2026) carries forward the hybrid Gated DeltaNet + sparse MoE architecture established in Qwen3-Next — per the qwen3.5 release.
• Qwen 3.6-Max (Apr 2, 2026) — succession of Qwen3-Max — Qwen 3.6-Max-Preview (April 2, 2026) is the next-generation proprietary closed-weights flagship, replacing Qwen3-Max in the Max product slot.
• Qwen3.6-35B-A3B (Apr 16, 2026) — succession of Qwen3.5 + Plus — Qwen3.6-35B-A3B (April 16, 2026) is the first open-weights Qwen3.6 release, a 35B-total/3B-active MoE — per HuggingFace Qwen/Qwen3.6-35B-A3B.
• Qwen3.6-27B (Apr 22, 2026) — new base of Qwen3.6-35B-A3B — Qwen3.6-27B (April 22, 2026) introduced the Gated DeltaNet + self-attention hybrid as a new dense base with Thinking Preservation — per qwen.ai/blog?id=qwen3.6-27b.
• Qwen3.7-Max (May 20, 2026) — succession of Qwen 3.6-Max — Qwen3.7-Max (May 20, 2026) is the closed-weights proprietary reasoning-agent flagship succeeding the 3.6-Max preview — per the Alibaba Cloud Summit announcement.
• Qwen3.7-Plus (May 31, 2026) — succession of Qwen3.7-Max — Qwen3.7-Plus (May 31, 2026) is the multimodal vision + language sibling to the text-only Qwen3.7-Max, together forming the Qwen3.7 generation announced at the May 20, 2026 Alibaba Cloud Summit. Alibaba's announcement describes Plus as retaining Max's coding / tool-use / reasoning strengths while adding image and video understanding, but does not disclose it as a fine-tune, post-training, or distillation of Qwen3.7-Max — per qwen.ai/blog?id=qwen3.7-plus and the Alibaba Cloud Bailian / Model Studio Recommended models page.
• Qwen3.8-Max (Aug 3, 2026) — succession of Qwen3.7-Max — Qwen3.8-Max was previewed at the World AI Conference in Shanghai on July 19, 2026 and reached general availability on August 3, 2026: a 2.4-trillion-parameter sparse-MoE model with hybrid attention, 95 billion active parameters, native vision-language input and a 1M-token context, billed on Qwen Cloud and Alibaba Cloud Model Studio as qwen3.8-max. Alibaba says it is 'built upon the architectural foundation of Qwen 3.5' — an architecture statement, not a base-model continuity claim — and discloses no fine-tune, post-training, or distillation relationship to Qwen3.7-Max, so the edge stays succession. Per alibabacloud.com/blog/qwen3-8-max-a-new-bar-for-coding-and-cowork_603421 and the docs.qwencloud.com/changelog/models entry for August 3, 2026.