A curated collection of landmark language models · Admission free
Collection Museum registry v1.0
Labs, companies, and developers that appear as model makers in this museum’s master catalog and on-site exhibits (alphabetical). Each tile shows v1.0 (registry checklist) and the earliest catalog row for that maker, or an exhibit date. Only the five premium provider catalog pages we ship (Google Gemini, OpenAI, Anthropic, xAI, Arcee AI) get a completed badge โ click v1.0 on those tiles to open the full-line museum page; everyone else stays registry-only until a catalog exists. Maintainers: sync with MODEL_MAKERS_CHECKLIST.md when adding rows.
Top-tier frontier models commonly offered as premium picks in early 2026 โ including Composer 2 as Cursor's own flagship, plus maximum tiers from Anthropic, OpenAI, xAI, Moonshot AI, Google DeepMind, Zhipu GLM-5, DeepSeek-V3.2, Xiaomi MiMo-V2-Pro, and Alibaba Qwen3-Max class lines.
Cursor's Q1 2026 flagship โ the default top-tier Composer for agentic multi-file work, deep codebase understanding, and long-running coding sessions inside the Cursor IDE.
Anthropic's strongest Claude tier for the hardest reasoning, long-horizon coding, and agentic work โ the flagship when latency and cost are secondary to quality. Listed as Thinking in Cursor-style menus.
OpenAI's broad flagship for multimodal tasks, tool use, and general intelligence โ the default "max quality" GPT line in first-quarter 2026 pickers.
xAI's Grok flagship for this era โ strong reasoning with a product focus on live data, speed, and personality-forward assistants.
Moonshot's Kimi line at K2.5 โ known for very long effective context and solid coding assistance; a first-class alternative in multi-model IDEs.
Google's Gemini 3.1 Pro โ the serious multimodal workhorse for complex prompts, documents, and vision-heavy workflows. Pickers often expose High and Low variants (quality vs. speed) and may badge them as new releases.
Zhipu’s flagship GLM-5 generation โ a large MoE stack (โ745B-class public framing) with very long context and a product focus on coding, agents, and Chinese-infrastructure inference paths. Positioned as a top-tier alternative in global and CN IDE pickers (often listed as GLM-5 next to other premium add-ons).
The V3.2 line continues DeepSeek’s MoE frontier with a balanced daily driver and a high-compute Speciale variant for deep reasoning โ plus open-weight releases and an emphasis on sparse attention and agentic tool-use training at scale.
MiMo-V2-Pro is Xiaomi’s top MiMo model for agentic workloads โ a trillion-scale MoE design with massive context (up to ~1M tokens in public materials), hybrid attention, and strong coding/agent benchmarks โ aimed at orchestration, planning, and long-horizon automation.
Alibaba’s Qwen3-Max family apex โ the Thinking variant adds test-time scaling and multi-pass refinement for hard math, code, and tool use. Served as a proprietary cloud flagship (e.g. DashScope) alongside open/smaller Qwen tiers in global stacks.
Companion and specialist models from the same generation โ Composer 1.5, Codex, Sonnet/Haiku, GPT-5.x variants, Gemini Flash tiers, and more. Snapshots: Qwen Code add-on billing row (its own card), Cursor context/comparison tables, Warp long model selector in the terminal, and other IDEs where noted. (Composer 2 lives in the flagship row above.)
IDE model menu (example snapshot)
Labels as shown in a typical Q1 2026 picker in Cursor and Google Antigravity โ New matches the in-app badge; indicates models that may show a caution in the UI.
Qwen Code โ add-on relative cost (example snapshot)
The add-on row in the Qwen Code UI in Cursor (Q1 2026). Relative cost multipliers are billing weights for these models (not vendor-published specs). List AโZ.
Warp โ long model selector (example snapshot)
Dropdown labels from a Q1 2026 Warp terminal agent session. US-hosted marks regional API endpoints. List AโZ.
Typical labels (AโZ)
Cursor โ context & comparison tables (example)
Context and score figures from in-app comparison UIs; they change as products update.
Context windows (Z.AI ยท MiniMax ยท OpenAI) โ example
| Model | Provider | Context (approx.) |
|---|---|---|
| GLM 4.6 | Z.AI | 200.0K |
| MiniMax M2.5 | MiniMax | 204.8K |
| MiniMax M2.1 | MiniMax | 204.8K |
| MiniMax M2.1 Lightning | MiniMax | 204.8K |
| GLM 4.7 | Z.AI | 200.0K |
| GLM 5 | Z.AI | 202.8K |
| GPT-5.3-Codex | OpenAI | 400.0K |
| GPT-5.2-Codex | OpenAI | 400.0K |
In-app comparison scores (example)
| Model | Score | Org | Context |
|---|---|---|---|
| Claude Opus 4.6 | 10 | Anthropic | 1.0M |
| Claude Opus 4.5 | 10 | Anthropic | 200.0K |
| Claude Sonnet 4.6 | 10 | Anthropic | 1.0M |
| Claude Sonnet 4.5 | 10 | Anthropic | 1.0M |
| GPT-5.4 | 10 | OpenAI | 1.0M |
| GPT-5.4 Pro | 10 | OpenAI | 1.0M |
| GPT-5.2 | 10 | OpenAI | 400.0K |
| GPT-5.1 | 10 | OpenAI | 400.0K |
| GPT-5.1-Codex-Max | 10 | OpenAI | 400.0K |
| Gemini 3 Pro Preview | 10 | 1.0M | |
| Gemini 3.1 Pro Preview | 10 | 1.0M | |
| GPT-5 mini | 9.9 | OpenAI | 400.0K |
| Claude Opus 4.1 | 9.8 | Anthropic | 200.0K |
| GPT-5 | 9.8 | OpenAI | 400.0K |
| Gemini 2.5 Pro (partial) | 9.6 | 1.0M |
OpenCode โ Select model (example snapshot)
From the OpenCode desktop appโs model picker (Jan 2026). Big Pickle is OpenCodeโs free Zen-tier coding model (often the default). Free = bundled no-cost rows; Recommended matches in-app hints for add-on providers. Official products: OpenCode Zen (curated agent models) and OpenCode Go (low-cost subscription).
Free models provided by OpenCode
Add more models from popular providers
Prior-generation Composer still common in menus โ balanced speed and reliability.
OpenAI Codex family tuned for IDE workflows โ strong inline and task completion.
Fast, capable Claude tier โ the usual sweet spot for everyday coding and agents. Same line as Thinking in Cursor pickers (extended reasoning budget).
Open-weight GPT-OSS family at 120B parameters โ Medium denotes the reasoning / compute tier in the menu, not the parameter count.
Earlier Opus generation โ still listed for compatibility and cost-sensitive depth work.
GPT-5 series model โ strong generalist when 5.4 is overkill or unavailable.
Smaller 5.4 variant โ faster turns and lower cost with shared family behavior.
Nano tier for lint-level fixes, tiny edits, and high-volume completions.
Latency-first Claude โ great for quick refactors, summaries, and chat throughput.
Spark variant โ optimized for snappy suggestions and lighter workloads.
Previous Sonnet โ familiar behavior for teams pinning older stacks.
Earlier Codex generation โ still appears in long-running project defaults.
5.1 family at Max width โ heavier coding sessions before 5.2/5.3.
GPT-5.1 checkpoint โ general assistant quality with broad tool compatibility.
Gemini 3 family speed tier โ high throughput for chat and multimodal bursts.
Compact Codex 5.1 โ ideal for tab completion and small-scope edits.
Sonnet 4 base โ predecessor to 4.5/4.6 feature tiers.
Budget GPT-5 class model โ everyday tasks without full 5.4 spend.
Widely deployed fast Gemini โ often the default "Flash" in multi-model lists.
HKChat (ๆธฏ่ฉฑ้) โ LLM-based assistant for Hong Kong citizens: local life Q&A, bilingual help, and preliminary legal orientation. Not a substitute for professional advice.
Public-facing chatbot positioned as a localized assistant: Cantonese, English, and Mandarin; Hong Kong–specific knowledge; Android, iOS, Windows, and HarmonyOS clients. Described in sources as built on HK-GNN V1 (Hong Kong Generative AI R&D Center) with Hong Kong data and values. Official entry points: hkchat.org, hkchat.org/ai-service, web app chat.hkchat.app.
The latest model from Anthropic โ and the one powering this very page right now. Claude Sonnet 4.6 (Thinking) is a fast, highly capable reasoning model with best-in-class coding, strong vision understanding, and agentic task execution. Built with a deep emphasis on helpfulness, harmlessness, and honesty. If you're reading this through a Cursor chat, hi.
Introduced "extended thinking" โ a visible chain-of-thought reasoning mode that lets Claude work through hard problems step by step before answering. Set new coding benchmarks at launch.
Trained on a cluster of 200,000 GPUs. Notable for "DeepSearch" โ a deep research mode that browses the web iteratively. Integrated with X (Twitter) for real-time context.
Achieved near-human scores on ARC-AGI, a benchmark designed to resist AI brute-force. The strongest reasoning model OpenAI has released, with compute-scaled thinking at inference time.
Open-source reasoning model that matched o1 on benchmarks at a fraction of the training cost. Caused significant market disruption at launch. Freely downloadable and runnable locally.
Google's fast flagship for 2025. Native multimodal โ handles text, images, audio, and video. Designed for agentic use cases and high-throughput applications with a generous free tier.
"Omni" โ the first model to handle text, image, and audio natively in one architecture. Introduced real-time voice conversation with human-like response speeds. Became ChatGPT's default model.
OpenAI's first publicly released reasoning model. Introduced the paradigm of spending more compute at inference time to "think" before responding โ enabling PhD-level performance on math and science.
Topped coding benchmarks for months at release. Introduced "computer use" โ the ability to control a desktop GUI like a human would. Became the go-to model for agentic coding workflows.
A mixture-of-experts model with 671B total parameters (37B active). Trained for $5.6M โ shocking efficiency compared to Western frontier models. Matched GPT-4o on benchmarks, fully open source.
Meta's openly released 70B model that outperformed the 405B Llama 3.1 on several benchmarks. A milestone for on-device and self-hosted AI โ runs on consumer hardware with quantization.
Introduced a breakthrough 1M token context window โ enough to hold entire codebases, books, or hours of video. Changed what "long context" meant for the industry.
Dense index of publicly notable LLM releases and open-weight checkpoints. This is not every private internal checkpoint; it follows the same broad canon as Wikipedia’s List of large language models, with extra gap-fill rows. Figures are often reported ranges or estimates, not audited specs.
| Model | Released | Developer | Parameters | Training data | Compute / cost | License | Notes |
|---|---|---|---|---|---|---|---|
| GPT-1 | Jun 2018 | OpenAI | 0.117B | - | - | MIT | First GPT; decoder-only transformer. |
| BERT | Oct 2018 | 0.34B | 3.3B words | - | Apache-2.0 | Encoder-only; highly influential. | |
| GPT-2 | Feb 2019 | OpenAI | 1.5B | ~10B tokens | - | MIT | Scaled LM generation. |
| T5 | Oct 2019 | 11B | 34B tokens | - | Apache-2.0 | Text-to-text transfer transformer. | |
| XLNet | Jun 2019 | 0.34B | 33B words | - | Apache-2.0 | Permutation LM. | |
| GPT-3 | May 2020 | OpenAI | 175B | 300B tokens | - | Proprietary | Few-shot learning at scale. |
| GPT-Neo | Mar 2021 | EleutherAI | 2.7B | 825 GiB | - | MIT | Open GPT-3-class alternative. |
| GPT-J | Jun 2021 | EleutherAI | 6B | The Pile | - | Apache-2.0 | Open autoregressive model. |
| Megatron-Turing NLG | Oct 2021 | Microsoft / NVIDIA | 530B | 338.6B tokens | - | Unreleased | Large-scale training on Selene. |
| Ernie 3.0 Titan | Dec 2021 | Baidu | 260B | 4 TB | - | Proprietary | Chinese LLM; Ernie Bot lineage. |
| Claude | Dec 2021 | Anthropic | 52B | 400B tokens | - | Proprietary | RLHF-style alignment focus. |
| GLaM | Dec 2021 | 1200B MoE | 1.6T tokens | - | Proprietary | Sparse MoE generalist. | |
| Gopher | Dec 2021 | Google DeepMind | 280B | 300B tokens | - | Proprietary | Later led to Chinchilla scaling insights. |
| LaMDA | Jan 2022 | 137B | 1.56T words | - | Proprietary | Dialog-specialized. | |
| GPT-NeoX | Feb 2022 | EleutherAI | 20B | 825 GiB | - | Apache-2.0 | Megatron-based. |
| Chinchilla | Mar 2022 | Google DeepMind | 70B | 1.4T tokens | - | Proprietary | Compute-optimal scaling law. |
| PaLM | Apr 2022 | 540B | 768B tokens | - | Proprietary | Pathways large model. | |
| OPT | May 2022 | Meta | 175B | 180B tokens | - | Non-commercial | Open replication effort + logbook. |
| YaLM 100B | Jun 2022 | Yandex | 100B | 1.7 TB | - | Apache-2.0 | ENโRU bilingual. |
| Minerva | Jun 2022 | 540B | 38.5B (math) | - | Proprietary | STEM reasoning; from PaLM. | |
| BLOOM | Jul 2022 | BigScience / HF | 175B | 350B tokens | - | RAIL | Multilingual open collaboration. |
| Galactica | Nov 2022 | Meta | 120B | 106B tokens | - | CC-BY-NC-4.0 | Scientific corpora. |
| AlexaTM | Nov 2022 | Amazon | 20B | 1.3T tokens | - | Proprietary | Seq2seq architecture. |
| Llama | Feb 2023 | Meta AI | 65B | 1.4T tokens | - | Research-only | Open-weights wave. |
| GPT-4 | Mar 2023 | OpenAI | Unknown | Unknown | - | Proprietary | Multimodal flagship era. |
| Cerebras-GPT | Mar 2023 | Cerebras | 13B | - | - | Apache-2.0 | Chinchilla-optimal training. |
| Falcon | Mar 2023 | TII | 40B | 1T tokens | - | Apache-2.0 | RefinedWeb + curated. |
| BloombergGPT | Mar 2023 | Bloomberg | 50B | 708B mixed | - | Unreleased | Finance-tuned. |
| PanGu-ฮฃ | Mar 2023 | Huawei | 1085B | 329B tokens | - | Proprietary | Very large dense/MoE stack. |
| OpenAssistant | Mar 2023 | LAION | 17B | 1.5T tokens | - | Apache-2.0 | Crowdsourced RLHF data. |
| Jurassic-2 | Mar 2023 | AI21 Labs | Unknown | - | - | Proprietary | API-first. |
| PaLM 2 | May 2023 | 340B | 3.6T tokens | - | Proprietary | Bard / workspace era. | |
| YandexGPT | May 2023 | Yandex | Unknown | - | - | Proprietary | Alice assistant. |
| Llama 2 | Jul 2023 | Meta AI | 70B | 2T tokens | - | Llama 2 license | Widespread finetunes. |
| Claude 2 | Jul 2023 | Anthropic | Unknown | - | - | Proprietary | Long-context Claude chat. |
| Granite 13B | Jul 2023 | IBM | 13B | - | - | Proprietary | watsonx.ai stack. |
| Mistral 7B | Sep 2023 | Mistral AI | 7.3B | - | - | Apache-2.0 | Efficient open weights. |
| Claude 2.1 | Nov 2023 | Anthropic | Unknown | - | - | Proprietary | ~200K token context. |
| Grok 1 | Nov 2023 | xAI | 314B | - | - | Apache-2.0 | Open-weight release; X integration. |
| Gemini 1.0 | Dec 2023 | Google DeepMind | Unknown | - | - | Proprietary | Multimodal family. |
| Mistral 8x7B | Dec 2023 | Mistral AI | 46.7B MoE | - | - | Apache-2.0 | MoE; strong benchmarks. |
| DeepSeek-LLM | Nov 2023 | DeepSeek | 67B | 2T tokens | - | DeepSeek License | EN + Chinese. |
| Phi-2 | Dec 2023 | Microsoft | 2.7B | 1.4T tokens | - | MIT | Textbook-quality data. |
| Gemini 1.5 | Feb 2024 | Google DeepMind | Unknown | - | - | Proprietary | 1M+ token context. |
| Gemini Ultra | Feb 2024 | Google DeepMind | Unknown | - | - | Proprietary | Benchmark-focused tier. |
| Gemma | Feb 2024 | Google DeepMind | 7B | 6T tokens | - | Gemma terms | Open-ish small models. |
| OLMo | Feb 2024 | Allen AI | 7B | 2T tokens | - | Apache-2.0 | Fully open pipeline. |
| Claude 3 | Mar 2024 | Anthropic | Unknown | - | - | Proprietary | Haiku / Sonnet / Opus. |
| DBRX | Mar 2024 | Databricks | 136B | 12T tokens | - | DBRX license | MoE; Mosaic training. |
| Mixtral 8x22B | Apr 2024 | Mistral AI | 141B MoE | - | - | Apache-2.0 | Larger MoE. |
| Phi-3 | Apr 2024 | Microsoft | 14B | 4.8T tokens | - | MIT | SLM marketing wave. |
| Qwen2 | Jun 2024 | Alibaba | 72B | 3T tokens | - | Qwen license | Multilingual. |
| DeepSeek-V2 | Jun 2024 | DeepSeek | 236B MoE | 8.1T tokens | - | DeepSeek License | Economic training. |
| Nemotron-4 | Jun 2024 | NVIDIA | 340B | 9T tokens | - | NVIDIA license | H100 cluster training. |
| Claude 3.5 | Jun 2024 | Anthropic | Unknown | - | - | Proprietary | Sonnet-led coding surge. |
| Llama 3.1 | Jul 2024 | Meta AI | 405B | 15.6T tokens | - | Llama 3 license | 405B flagship open-ish. |
| Grok-2 | Aug 2024 | xAI | Unknown | - | - | xAI license | Later Grok 2.5 source-available. |
| OpenAI o1 | Sep 2024 | OpenAI | Unknown | - | - | Proprietary | Explicit reasoning model. |
| Mistral Large | Nov 2024 | Mistral AI | 123B | - | - | Mistral Research | API flagship. |
| Pixtral | Nov 2024 | Mistral AI | 123B | - | - | Mistral Research | Multimodal. |
| OLMo 2 | Nov 2024 | Allen AI | 32B | 6.6T tokens | - | Apache-2.0 | Open research LM. |
| Phi-4 | Dec 2024 | Microsoft | 14B | 9.8T tokens | - | MIT | SLM continued. |
| DeepSeek-V3 | Dec 2024 | DeepSeek | 671B MoE | 14.8T tokens | - | MIT | Cost-shock open weights. |
| Amazon Nova | Dec 2024 | Amazon | Unknown | - | - | Proprietary | Micro / Lite / Pro. |
| DeepSeek-R1 | Jan 2025 | DeepSeek | 671B | RL only | - | MIT | Reasoning from base. |
| Qwen2.5 | Jan 2025 | Alibaba | 72B | 18T tokens | - | Qwen license | Dense + MoE lineup. |
| MiniMax-Text-01 | Jan 2025 | MiniMax | 456B | 4.7T tokens | - | MiniMax license | Long-context focus. |
| Gemini 2.0 | Feb 2025 | Google DeepMind | Unknown | - | - | Proprietary | Flash / Flash-Lite / Pro. |
| Claude 3.7 | Feb 2025 | Anthropic | Unknown | - | - | Proprietary | Sonnet + extended thinking. |
| GPT-4.5 | Feb 2025 | OpenAI | Unknown | - | - | Proprietary | Largest non-reasoning GPT then. |
| Grok 3 | Feb 2025 | xAI | Unknown | - | - | Proprietary | Massive compute claims. |
| Gemini 2.5 | Mar 2025 | Google DeepMind | Unknown | - | - | Proprietary | Flash / Flash-Lite / Pro. |
| Llama 4 | Apr 2025 | Meta AI | 400B | 40T tokens | - | Llama 4 license | Multimodal natively. |
| OpenAI o3 / o4-mini | Apr 2025 | OpenAI | Unknown | - | - | Proprietary | Reasoning stack. |
| Qwen3 | Apr 2025 | Alibaba | 235B | 36T tokens | - | Apache-2.0 | Many sizes down to 0.6B. |
| Claude 4 | May 2025 | Anthropic | Unknown | - | - | Proprietary | Sonnet + Opus refresh. |
| Sarvam-M | May 2025 | Sarvam AI | 24B | - | - | Apache-2.0 | India-focused reasoning. |
| Grok 4 | Jul 2025 | xAI | Unknown | - | - | Proprietary | Frontier Grok line. |
| Param-1 | Jul 2025 | BharatGen | 2.9B | 5T tokens | - | Unknown | Indic languages. |
| GLM-4.5 | Jul 2025 | Zhipu AI | 355B MoE | 22T tokens | - | MIT | 335B / 106B sizes. |
| GPT-OSS | Aug 2025 | OpenAI | 117B | - | - | Apache-2.0 | 20B + 120B open weights. |
| Claude 4.1 | Aug 2025 | Anthropic | Unknown | - | - | Proprietary | Opus refresh. |
| GPT-5 | Aug 2025 | OpenAI | Unknown | - | - | Proprietary | Mini / nano / full family. |
| DeepSeek-V3.1 | Aug 2025 | DeepSeek | 671B | 15.6T+ tokens | - | MIT | Hybrid thinking modes. |
| Apertus | Sep 2025 | ETH / EPFL | 70B | 15T tokens | - | Apache-2.0 | EU AI Act positioning. |
| Claude Sonnet 4.5 | Sep 2025 | Anthropic | Unknown | - | - | Proprietary | Coding + agents. |
| DeepSeek-V3.2-Exp | Sep 2025 | DeepSeek | 685B | - | - | MIT | DSA sparse attention. |
| GLM-4.6 | Sep 2025 | Zhipu AI | 357B | - | - | Apache-2.0 | Open flagship coding. |
| Gemini 3 | Nov 2025 | Google DeepMind | Unknown | - | - | Proprietary | Deep Think / Pro tiers. |
| Olmo 3 | Nov 2025 | Allen AI | 32B | 5.9T tokens | - | Apache-2.0 | 7B + 32B reasoning. |
| Claude Opus 4.5 | Nov 2025 | Anthropic | Unknown | - | - | Proprietary | Largest Claude then. |
| GPT-5.2 | Dec 2025 | OpenAI | Unknown | - | - | Proprietary | Reasoning + pro workloads. |
| GLM-4.7 | Dec 2025 | Zhipu AI | 355B MoE | - | - | Apache-2.0 | MoE SOTA coding claims. |
| Qwen3-Max-Thinking | Jan 2026 | Alibaba | Unknown | - | - | Proprietary | Adaptive tool use. |
| Kimi K2.5 | Jan 2026 | Moonshot AI | 1040B MoE | 15T tokens | - | Modified MIT | 32B active; multimodal. |
| Claude Opus 4.6 | Feb 2026 | Anthropic | Unknown | - | - | Proprietary | Frontier Claude. |
| GPT-5.3-Codex | Feb 2026 | OpenAI | Unknown | - | - | Proprietary | Agentic coding line. |
| GPT-5.2-Codex | Feb 2026 | OpenAI | Unknown | - | - | Proprietary | Codex family; large context in tools. |
| GLM-5 | Feb 2026 | Zhipu AI | 754B | - | - | MIT | DSA; 200K context. |
| Qwen-Coder-Qoder-1.0 | 2026 | Alibaba | Unknown | - | - | Proprietary | IDE add-on label; Qwen coding line. |
| Qwen3.5-Plus | 2026 | Alibaba | Unknown | - | - | Proprietary | Plus tier in model pickers. |
| MiniMax M2.1 | 2025 | MiniMax | Unknown | - | - | Proprietary | M2 API family; ~205K context. |
| MiniMax M2.1 Lightning | 2025 | MiniMax | Unknown | - | - | Proprietary | Speed-focused M2.1 variant. |
| MiniMax M2.5 | 2025 | MiniMax | Unknown | - | - | Proprietary | Long-context M2 line; ~205K. |
| MiniMax-M2.7 | 2026 | MiniMax | Unknown | - | - | Proprietary | Add-on tier in Cursor-style menus. |
| Param-2 | Feb 2026 | BharatGen | 17B MoE | ~22T tokens | - | Unknown | More Indic langs. |
| Sarvam-1 | Feb 2026 | Sarvam AI | 105B MoE | ~12T tokens | - | Apache-2.0 | India foundation model. |
| GPT-5.4 | Mar 2026 | OpenAI | Unknown | - | - | Proprietary | Thinking + Pro variants. |
| GPT-5.4 Pro | Mar 2026 | OpenAI | Unknown | - | - | Proprietary | Pro tier; ~1M context in pickers. |
| GPT-5.1 Codex | 2026 | OpenAI | Unknown | - | - | Proprietary | Codex line in IDE menus. |
| GPT-5.1 Codex Max | 2026 | OpenAI | Unknown | - | - | Proprietary | Top Codex tier in pickers. |
| Grok 4.20 | 2026 | xAI | Unknown | - | - | Proprietary | Picker label; see Grok 4 article. |
| Gemini 3.1 Pro | 2026 | Google DeepMind | Unknown | - | - | Proprietary | Pro tier naming in tools. |
| Gemini 3 Pro Preview | 2026 | Google DeepMind | Unknown | - | - | Proprietary | Preview label in model lists. |
| Gemini 3.1 Pro Preview | 2026 | Google DeepMind | Unknown | - | - | Proprietary | Preview label in model lists. |
| Gemini 2.5 Pro (partial) | 2025 | Google DeepMind | Unknown | - | - | Proprietary | Pro tier; partial availability. |
| Claude Opus 4.1 | Aug 2025 | Anthropic | Unknown | - | - | Proprietary | Opus tier naming in pickers. |
| Composer 2 | 2026 | Cursor / Anysphere | - | - | - | Proprietary | IDE agent flagship (product). |
| Codex | Aug 2021 | OpenAI | 12B | - | - | Proprietary | Code fine-tune of GPT-3 lineage. |
| InstructGPT | Jan 2022 | OpenAI | 175B | - | - | Proprietary | RLHF alignment showcase. |
| GPT-4 Turbo | Nov 2023 | OpenAI | Unknown | - | - | Proprietary | 128K context; cheaper GPT-4 class. |
| YandexGPT 2 | Sep 2023 | Yandex | Unknown | - | - | Proprietary | Alice assistant update. |
| Llama 3 | Apr 2024 | Meta AI | 8-70B | 15T tokens | - | Llama 3 license | Dense family before 3.1. |
| GPT-4o | May 2024 | OpenAI | Unknown | - | - | Proprietary | Omni multimodal flagship. |
| YandexGPT 3 Pro | Mar 2024 | Yandex | Unknown | - | - | Proprietary | Alice chatbot. |
| YandexGPT 3 Lite | May 2024 | Yandex | Unknown | - | - | Proprietary | Alice chatbot. |
| Fugaku-LLM | May 2024 | Fujitsu / Titech et al. | 13B | 380B tokens | - | Fugaku terms | CPU-trained on Fugaku. |
| Chameleon | May 2024 | Meta AI | 34B | 4.4T tokens | - | Non-commercial | Early-token fusion multimodal. |
| o1-mini | Sep 2024 | OpenAI | Unknown | - | - | Proprietary | Smaller reasoning model. |
| YandexGPT 4 Lite/Pro | Oct 2024 | Yandex | Unknown | - | - | Proprietary | Alice chatbot. |
| Llama 3.2 | Sep 2024 | Meta AI | 1-90B | - | - | Llama 3.2 license | Vision + text stack. |
| Llama 3.3 70B | Dec 2024 | Meta AI | 70B | - | - | Llama 3.3 license | Strong 70B after 405B 3.1. |
| DeepSeek-V3-0324 | Mar 2025 | DeepSeek | 671B | 14.8T+ ext. | - | MIT | V3 refresh checkpoint. |
| YandexGPT 5 Lite Pretrain/Pro | Feb 2025 | Yandex | Unknown | - | - | Proprietary | Alice Neural Network. |
| YandexGPT 5 Lite Instruct | Mar 2025 | Yandex | Unknown | - | - | Proprietary | Alice Neural Network. |
| Gemini 2.5 Flash | Mar 2025 | Google DeepMind | Unknown | - | - | Proprietary | Fast tier in 2.5 family. |
| YandexGPT 5.1 Pro | Aug 2025 | Yandex | Unknown | - | - | Proprietary | Alice Neural Network. |
| Alice AI LLM 1.0 | Oct 2025 | Yandex | Unknown | - | - | Proprietary | Alice AI chatbot. |
| Claude Haiku 4.5 | Oct 2025 | Anthropic | Unknown | - | - | Proprietary | Fast Claude tier (companion lineup). |