AI models released in November 2023

27 releases from 15 labs in November 2023, including Qwen-72B and Qwen-1.8B, Qwen-Audio and Seamless (SeamlessM4T v2, SeamlessExpressive, SeamlessStreaming). Every entry is dated to the lab's official announcement and links to it.

Releases
27
Labs
15
Open weights
13

November 2023

27 releases
  1. Alibaba QwenLLMOpen weights

    Qwen-72B and Qwen-1.8B

    Qwen-72B is the largest first-generation Qwen model, trained on 3T tokens with 32K context, released with the compact Qwen-1.8B and chat versions of both.

    Modalities
    text → text
    Context
    32K context
    Parameters
    1.8B, 72B
    License
    Tongyi Qianwen License Agreement (1.8B uses the Tongyi Qianwen Research License Agreement)
    At launch
    Open weights
    • Qwen-72B trained on 3.0T tokens; Qwen-1.8B on 2.2T
    • Both support 32K context
    • Released on ModelScope and Hugging Face
    Source: Qwen (GitHub)
  2. Alibaba QwenMultimodalOpen weights

    Qwen-Audio

    Qwen-Audio is Qwen's audio-language model that accepts diverse audio and text input, released with the instruction-tuned Qwen-Audio-Chat. Its technical paper was published on November 15, 2023.

    Modalities
    text, audio → text
    License
    Tongyi Qianwen License Agreement
    At launch
    Open weights
    • Checkpoints for Qwen-Audio and Qwen-Audio-Chat on ModelScope and Hugging Face
    • Commercial use allowed under its license
    • Followed by Qwen2-Audio in August 2024
    Source: Qwen-Audio (GitHub)
  3. MetaSpeechOpen weights

    Seamless (SeamlessM4T v2, SeamlessExpressive, SeamlessStreaming)

    Seamless is a suite of four translation models built on SeamlessM4T v2, adding expression-preserving speech translation and streaming translation with about two seconds of latency.

    Modalities
    text, audio → text, audio
    Parameters
    2.3B (SeamlessM4T v2 Large), 2.5B (SeamlessStreaming)
    License
    CC-BY-NC-4.0 (SeamlessM4T v2, SeamlessStreaming); Seamless License (SeamlessExpressive)
    At launch
    Open weights
    • SeamlessExpressive preserves expression across English, Spanish, German, French, Italian and Chinese
    • SeamlessStreaming supports speech recognition and speech-to-text translation for nearly 100 languages
    • All four models released publicly for researchers
    Source: Introducing a suite of AI language translation models that preserve expression and improve streaming
  4. AmazonImageProprietary

    Amazon Titan Image Generator G1

    Amazon Titan Image Generator G1 is Amazon's first image generation model, launched in public preview with text-to-image generation from English prompts, text-prompted editing, inpainting and outpainting.

    Modalities
    text, image → image
    At launch
    API
    • Launched in public preview in US East (N. Virginia) and US West (Oregon)
    • All generated images contain an invisible watermark
    • Supports fine-tuning on proprietary data
    Source: Amazon Titan Image Generator, Multimodal Embeddings, and Text models are now available in Amazon Bedrock
  5. AmazonEmbeddingProprietary

    Amazon Titan Multimodal Embeddings G1

    Amazon Titan Multimodal Embeddings G1 generates embeddings from text, images or both together, enabling multimodal search and recommendation applications in Amazon Bedrock.

    Modalities
    text, image → embedding
    Context
    0K context
    At launch
    API
    • Accepts text, image, or text and image together, with text limited to 128 tokens
    • Defaults to 1,024-dimension vectors, with smaller sizes available
    • Generally available in US East (N. Virginia) and US West (Oregon)
    Source: Amazon Titan Multimodal Embeddings foundation model now generally available in Amazon Bedrock
  6. AmazonLLMProprietary

    Amazon Titan Text G1 Express

    Amazon Titan Text G1 Express is Amazon's general-purpose text generation model for open-ended generation, chat and retrieval-augmented generation, generally available in Amazon Bedrock after a limited preview of the Titan Text family.

    Modalities
    text → text
    Context
    8K context
    At launch
    API
    • Maximum context length of 8,192 tokens
    • Optimized for English, with 100+ additional languages in preview
    • Titan Text was first previewed with select customers in April 2023
    Source: Amazon Titan Text models, Express and Lite, now generally available in Amazon Bedrock
  7. AmazonSmall modelProprietary

    Amazon Titan Text G1 Lite

    Amazon Titan Text G1 Lite is the fastest model in Amazon's Titan Text family, a smaller, cost-effective option aimed at English summarization and copywriting tasks and well suited to fine-tuning.

    Modalities
    text → text
    Context
    4K context
    At launch
    API
    • Maximum context length of 4,096 tokens
    • Aimed at English-language tasks
    • Launched in five AWS Regions including US East (N. Virginia) and Europe (Frankfurt)
    Source: Amazon Titan Text models, Express and Lite, now generally available in Amazon Bedrock
  8. DeepSeekLLMOpen weights

    DeepSeek LLM

    DeepSeek LLM is a general-purpose language model family released in 7B and 67B sizes as base and chat models. The models were trained from scratch on 2T English and Chinese tokens.

    Modalities
    text → text
    Context
    4K context
    Parameters
    7B, 67B
    License
    DeepSeek Model License
    At launch
    Open weights
    • Pre-trained with a 4,096-token sequence length
    • DeepSeek also released intermediate base model checkpoints for research
    • Supports commercial use under the DeepSeek Model License
    Source: DeepSeek LLM: Let there be answers (GitHub)
  9. PerplexityLLMProprietary

    pplx-7b-online and pplx-70b-online

    Two online LLMs fine-tuned from Mistral 7B and Llama 2 70B to answer with snippets from Perplexity's own search index, giving API developers responses grounded in current web information.

    Modalities
    text → text
    Parameters
    pplx-7b-online: 7B (Mistral 7B base); pplx-70b-online: 70B (Llama 2 70B base)
    At launch
    API, Consumer app
    • Available through pplx-api, which moved from beta to general release
    • Also accessible in Perplexity Labs, the company's LLM playground
    • Offline pplx-7b-chat and pplx-70b-chat models left alpha in the same release
    Source: Introducing PPLX Online LLMs
  10. Stability AIImageOpen weights

    SDXL Turbo

    Real-time text-to-image model distilled from SDXL 1.0 with Adversarial Diffusion Distillation, cutting the required sampling steps from 50 to one while keeping high fidelity.

    Modalities
    text → image
    License
    Stability AI Community License
    At launch
    Open weights, Consumer app
    • Single-step image generation
    • Initially released under a non-commercial research license
    • Beta real-time demo on Clipdrop
    Source: Introducing SDXL Turbo: A Real-Time Text-to-Image Generation Model
  11. 01.AILLMOpen weights

    Yi-34B-Chat and Yi-6B-Chat

    01.AI open-sourced Yi-34B-Chat and Yi-6B-Chat, chat models fine-tuned from its Yi base models, together with 4-bit AWQ and 8-bit GPTQ quantized versions for consumer-grade GPUs.

    Modalities
    text → text
    Context
    4K context
    Parameters
    6B, 34B
    License
    Apache-2.0
    At launch
    Open weights
    • Chat models trained with supervised fine-tuning
    • Released with four quantized variants
    • Yi Series Models Community License updated to v2.1 the same day; weights are now under Apache 2.0
    Source: GitHub - 01-ai/Yi: A series of large language models trained from scratch by developers @01-ai
  12. ElevenLabsSpeechProprietary

    Speech to Speech (Voice Changer)

    Voice conversion model, originally called speech-to-speech, that makes a recording sound as if spoken by another voice while preserving emotion, timing, pacing and pronunciation.

    Modalities
    audio → audio
    At launch
    Consumer app, API
    • Preserves the original performance
    • Added to Speech Synthesis
    • Multilingual version eleven_multilingual_sts_v2 followed
    Source: Speech to Speech
  13. AnthropicLLMProprietary

    Claude 2.1

    Claude 2.1 doubled Claude's context window to 200,000 tokens and, according to Anthropic, made fewer false statements than Claude 2.0. It also added system prompts and a beta tool use feature.

    Modalities
    text → text
    Context
    200K context
    At launch
    API, Consumer app
    • API model ID claude-2.1
    • 200,000-token context window
    • Powers claude.ai free and Pro tiers; 200K context use on claude.ai reserved for Claude Pro
    Source: Introducing Claude 2.1
  14. Stability AIVideoOpen weights

    Stable Video Diffusion

    Stability AI's first open video model, released in research preview as two image-to-video models generating 14 and 25 frames at customizable frame rates between 3 and 30 frames per second.

    Modalities
    image → video
    License
    Stability AI Community License
    At launch
    Open weights, API
    • Code on GitHub and weights on Hugging Face
    • API access added on December 20, 2023
    • Current Hugging Face license is the Stability AI Community License
    Source: Introducing Stable Video Diffusion
  15. MicrosoftLLMOpen weights

    Orca 2

    Orca 2 is a pair of Microsoft Research models fine-tuned from Llama 2 on synthetic data to teach smaller models different reasoning strategies, such as step-by-step processing and direct answering.

    Modalities
    text → text
    Parameters
    7B, 13B
    License
    Microsoft Research License
    At launch
    Open weights
    • Released in 7B and 13B sizes
    • Fine-tuned from Llama 2 base models
    • Not trained with RLHF for safety, per Microsoft Research
    Source: Orca 2: Teaching Small Language Models How to Reason
  16. GoogleMusicProprietary

    Lyria

    Lyria is Google DeepMind's music generation model, announced with YouTube, built to generate music with instrumentals and vocals, perform transformation and continuation tasks and give finer control over style and performance.

    Modalities
    text, audio → audio
    At launch
    Consumer app
    • Trialed in Dream Track, a YouTube Shorts experiment for a limited set of creators
    • Dream Track generated 30-second soundtracks with lyrics, backing track and an AI voice in a participating artist's style
    • All content published by Lyria was watermarked with SynthID
    Source: Transforming the future of music creation
  17. NVIDIALLMOpen weights

    Nemotron-3 8B

    Family of 8B enterprise language models released as NVIDIA AI Foundation Models, with base, chat and question-answering variants for building custom chatbots and copilots.

    Modalities
    text → text
    Parameters
    8B
    At launch
    Open weights, Cloud platforms
    • Chat variants tuned with SFT, RLHF and SteerLM
    • Available in the Azure AI model catalog, Hugging Face and NVIDIA's model hub
    • Includes a Nemotron-3-8B-QA model
    Source: NVIDIA AI Foundation Models: Build Custom Enterprise Chatbots and Co-Pilots with Production-Ready LLMs | NVIDIA Technical Blog
  18. ElevenLabsSpeechProprietary

    Eleven Turbo v2

    Low-latency English speech model generating speech at about 400ms latency, over twice as fast as ElevenLabs' V1 models with quality on par with Multilingual v2.

    Modalities
    text → audio
    At launch
    API, Consumer app
    • About 400ms latency
    • Added mulaw 8kHz output for VoIP
    • Later outclassed by Flash v2
    Source: Turbo v2 is here
  19. OpenAILLMProprietary

    gpt-3.5-turbo-1106

    An updated GPT-3.5 Turbo model announced at OpenAI DevDay, supporting a 16K context window by default along with improved instruction following, JSON mode, and parallel function calling.

    Modalities
    text → text
    Context
    16K context
    At launch
    API
    • API model ID gpt-3.5-turbo-1106
    • 16,385-token context window per the GPT-3.5 Turbo docs model page
    • Fine-tuning supports 16K context with the new model
    • Older gpt-3.5-turbo-0613 stayed accessible until June 13, 2024
    Source: New models and developer products announced at DevDay
  20. OpenAILLMProprietary

    GPT-4 Turbo and GPT-4 Turbo with Vision

    GPT-4 Turbo is a preview of the next generation of GPT-4 with a 128K context window, knowledge of events up to April 2023, JSON mode, and improved function calling, plus a vision variant that accepts image inputs.

    Modalities
    text, image → text
    Context
    125K context
    At launch
    API
    • Preview model IDs gpt-4-1106-preview and gpt-4-vision-preview
    • Supports parallel function calling, JSON mode, and the seed parameter
    • Generally available with vision on Apr 9, 2024 as gpt-4-turbo-2024-04-09 (API changelog)
    • Updated preview gpt-4-0125-preview followed on Jan 25, 2024
    Source: New models and developer products announced at DevDay
  21. OpenAISpeechProprietary

    TTS (tts-1 and tts-1-hd)

    OpenAI launched a text-to-speech API with a new TTS model offered in two variants, tts-1 optimized for real-time use and tts-1-hd optimized for quality, with six preset voices to choose from.

    Modalities
    text → audio
    At launch
    API
    • Model IDs tts-1 and tts-1-hd in the Audio API speech endpoint
    • Six preset voices at launch
    • Announced at OpenAI DevDay alongside GPT-4 Turbo and the DALL·E 3 API
    Source: New models and developer products announced at DevDay
  22. OpenAISpeechOpen weights

    Whisper large-v3

    Whisper large-v3 is the next version of OpenAI's open-source automatic speech recognition model, released at DevDay with improved performance across languages, a 128-bin mel spectrogram input, and a new Cantonese language token.

    Modalities
    audio → text
    Parameters
    1550M
    License
    MIT
    At launch
    Open weights
    • Released via the openai/whisper GitHub repo and on Hugging Face as openai/whisper-large-v3
    • Same architecture as large and large-v2, with 1550M parameters
    • Trained on 1 million hours of weakly labeled and 4 million hours of pseudo-labeled audio
    • GitHub README states MIT for code and weights; the Hugging Face card is tagged apache-2.0
    Source: New models and developer products announced at DevDay
  23. 01.AILLMOpen weights

    Yi-6B-200K and Yi-34B-200K

    01.AI released Yi-6B-200K and Yi-34B-200K, base models with the same sizes as the original Yi release but a context window extended to 200K tokens.

    Modalities
    text → text
    Context
    200K context
    Parameters
    6B, 34B
    License
    Apache-2.0
    At launch
    Open weights
    • 200K tokens is roughly 400,000 Chinese characters, per 01.AI
    • Yi-34B-200K long-text capability was enhanced on March 7, 2024 with 5B tokens of long-context pre-training
    • Bilingual English and Chinese base models
    Source: GitHub - 01-ai/Yi: A series of large language models trained from scratch by developers @01-ai
  24. xAILLMProprietary

    Grok (Grok-1)

    Grok is xAI's first chatbot, powered by the Grok-1 language model developed over four months, with real-time knowledge through the X platform. It launched as an early beta for a limited number of US users.

    Modalities
    text → text
    Context
    8K context
    At launch
    Consumer app
    • Grok-1 model card lists an 8,192-token context length
    • Succeeded the 33B-parameter Grok-0 prototype
    • Offered first via a waitlist; company timeline lists Grok on X on December 7, 2023
    Source: Announcing Grok | SpaceXSI
  25. DeepSeekCodingOpen weights

    DeepSeek Coder

    DeepSeek Coder is a series of open-weights code models trained from scratch on 2T tokens of 87% code and 13% natural language. Base models support project-level code completion and infilling.

    Modalities
    text → text
    Context
    16K context
    Parameters
    1.3B, 5.7B, 6.7B, 33B
    License
    DeepSeek Model License
    At launch
    Open weights
    • Sizes of 1.3B, 5.7B, 6.7B, and 33B, with Base and Instruct models
    • Pre-trained with a 16K window and a fill-in-the-blank task
    • Covers more than 80 programming languages; supports commercial use
    Source: DeepSeek Coder: Let the Code Write Itself (GitHub)
  26. CohereEmbeddingProprietary

    Embed v3

    Embedding model that judges both how well a document matches a query's topic and the quality of its content, aimed at search and RAG over noisy real-world data.

    Modalities
    text → embedding
    At launch
    API
    • Uses compression-aware training
    • English, multilingual and light variants published
    • Later extended to image inputs as multimodal Embed 3
    Source: Introducing Embed v3
  27. 01.AILLMOpen weights

    Yi-6B and Yi-34B

    Yi-6B and Yi-34B were 01.AI's first public release: bilingual English and Chinese base language models trained from scratch on 3T tokens.

    Modalities
    text → text
    Context
    4K context
    Parameters
    6B, 34B
    License
    Apache-2.0
    At launch
    Open weights
    • Trained with 4K sequence length and extendable to 32K at inference time
    • Training data runs up to June 2023
    • Originally released under the Yi Series Models Community License; weights are now distributed under Apache 2.0
    Source: GitHub - 01-ai/Yi: A series of large language models trained from scratch by developers @01-ai

Cite and reuse this data

These release records are free to reuse under CC BY 4.0, including commercially. Credit The AI Engineering and link to this page's URL.

Dates are the labs' official announcement dates, and every record links to its source. List fields in the CSV are separated by " | ".

Citation

The AI Engineering Editors. "AI models released in November 2023." The AI Engineering, updated Nov 30, 2023. https://theaiengineering.com/releases/2023/11

HTML link

<a href="https://theaiengineering.com/releases/2023/11">AI models released in November 2023</a> (The AI Engineering)
Full dataset
1,167 releases from 55 labs
Coverage
to
Updated
License
CC BY 4.0
Publisher
The AI Engineering

Archive