AI models released in April 2024
21 releases from 13 labs in April 2024, including Amazon Titan Text Embeddings V2, Qwen1.5-110B and Snowflake Arctic. Every entry is dated to the lab's official announcement and links to it.
- Releases
- 21
- Labs
- 13
- Open weights
- 16
April 2024
21 releasesAmazon Titan Text Embeddings V2
Amazon Titan Text Embeddings V2 is Amazon's second-generation text embedding model, optimized for retrieval-augmented generation, with selectable output dimensions, default normalization and pre-training on more than 100 languages plus code.
- Modalities
- text → embedding
- Context
- 8K context
- At launch
- API
- Accepts up to 8,192 input tokens
- Output dimensions of 256, 512 or 1,024
- Launched in US East (N. Virginia) and US West (Oregon)
Qwen1.5-110B
Qwen1.5-110B is the first 100B-plus model in the Qwen1.5 series, built with the same Transformer decoder architecture as other Qwen1.5 models.
- Modalities
- text → text
- Context
- 32K context
- Parameters
- 110B
- License
- Tongyi Qianwen License Agreement
- At launch
- Open weights
- 32K-token context length
- Multilingual, including English, Chinese, French, Spanish, German, Russian, Korean, and Japanese
- Weights under the Tongyi Qianwen License Agreement
Snowflake Arctic
Enterprise-focused open language model with a dense-MoE hybrid architecture that combines a 10B dense transformer with a residual 128-expert MoE layer.
- Modalities
- text → text
- Parameters
- Dense-MoE hybrid: 480B total, 17B active
- License
- Apache-2.0
- At launch
- Open weights
- Uses top-2 gating across 128 fine-grained experts
- Base and instruct checkpoints released under Apache 2.0
- Data recipes and research insights open-sourced
OpenELM
OpenELM is Apple's family of small open language models that use layer-wise scaling to allocate parameters across transformer layers, released with pretrained and instruction-tuned versions.
- Modalities
- text → text
- Parameters
- 270M, 450M, 1.1B, 3B
- License
- Apple Machine Learning Research Model License
- At launch
- Open weights
- Pretrained on about 1.8 trillion tokens of public data
- Release includes training and evaluation framework, logs and checkpoints
- Includes code to convert models to MLX for Apple devices
Phi-3-mini
Phi-3-mini is a 3.8B instruction-tuned Microsoft small language model and the first of the Phi-3 family, offered in 4K and 128K context variants for on-device, latency-sensitive and cost-constrained uses.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 3.8B
- License
- MIT
- At launch
- Open weights, API, Cloud platforms
- Available in 4K and 128K token context variants
- Released on Azure AI Studio, Hugging Face and Ollama, and as an NVIDIA NIM microservice
- Safety post-training included RLHF and red-teaming
Llama 3
Llama 3 is Meta's third-generation open model family, released as 8B and 70B pretrained and instruction-tuned models trained on over 15 trillion tokens.
- Modalities
- text → text
- Context
- 8K context
- Parameters
- 8B, 70B
- License
- Meta Llama 3 Community License
- At launch
- Open weights, Cloud platforms
- Trained on sequences of 8,192 tokens
- Released with Llama Guard 2, Code Shield and CyberSec Eval 2
- Meta said models over 400B parameters were still in training
Llama Guard 2
Llama Guard 2 is an 8B Llama 3 based input and output safeguard model released with Llama 3, using the MLCommons hazard taxonomy.
- Modalities
- text → text
- Parameters
- 8B
- License
- Meta Llama 3 Community License
- At launch
- Open weights
- Predicts safety labels on 11 of the 13 MLCommons hazard categories
- Can be fine-tuned for different applications
Mixtral 8x22B
Larger sparse mixture-of-experts model released under Apache 2.0, with native function calling, strong math and coding skills and a 64K-token context window.
- Modalities
- text → text
- Context
- 64K context
- Parameters
- MoE: 141B total, 39B active
- License
- Apache-2.0
- At launch
- Open weights, API
- Fluent in English, French, Italian, German and Spanish
- Base and instruct checkpoints released
- API model ID open-mixtral-8x22b
OLMo 1.7 7B
Updated 7B OLMo model with a longer context window, a new two-stage training procedure and improved data, released with the Dolma 1.7 dataset.
- Modalities
- text → text
- Context
- 4K context
- Parameters
- 7B
- License
- Apache-2.0
- At launch
- Open weights
- Context length raised from 2,048 to 4,096 tokens
- Trained on the new Dolma 1.7 data mix
- Released under Apache 2.0
Arctic Embed
Snowflake's family of open text embedding models optimized for retrieval quality at each size, released on Hugging Face under Apache 2.0.
- Modalities
- text → embedding
- License
- Apache-2.0
- At launch
- Open weights
- Released in xs, s, m, m-long and l sizes
- A technical report followed in May 2024
- Arctic Embed M v1.5 added compressible vectors in July 2024
CodeQwen1.5
CodeQwen1.5-7B is a specialized code model built on the Qwen1.5 language model and pre-trained on around 3T tokens of code-related data.
- Modalities
- text → text
- Context
- 64K context
- Parameters
- 7B
- License
- Tongyi Qianwen License Agreement
- At launch
- Open weights
- Supports 92 programming languages
- Processes up to 64K tokens of context
- Weights under the Tongyi Qianwen License Agreement
Idefics2
General multimodal model that takes arbitrary sequences of text and images, improving on Idefics1 with a smaller size, an open license and stronger OCR.
- Modalities
- text, image → text
- Parameters
- 8B
- License
- Apache-2.0
- At launch
- Open weights
- 8B parameters under Apache 2.0
- Improved optical character recognition
- Base and instruct versions released
Reka Core
Reka's largest model at launch, a multimodal language model trained from scratch that accepts text, image, video and audio input and was offered via API, on-premises or on-device deployment.
- Modalities
- text, image, video, audio → text
- At launch
- API, Consumer app, Enterprise
- 128K context window
- Pretrained on text from 32 languages
- Launch partners included Snowflake (Cortex), Oracle and AI Singapore
- Technical report published alongside the launch
Rerank 3
Reranking foundation model for enterprise search and RAG that handles longer documents and semi-structured data such as emails, JSON, code and tables.
- Modalities
- text → score
- Context
- 4K context
- At launch
- API
- 4K-token context length
- Covers 100+ languages
- Improved latency and lower total cost of ownership
CodeGemma
CodeGemma is a family of Gemma-based open code models for code completion, generation and chat. It includes a 7B pretrained model, a 7B instruction-tuned model and a 2B model for fast local code completion.
- Modalities
- text → text
- Parameters
- 2B, 7B
- License
- Gemma Terms of Use
- At launch
- Open weights, Cloud platforms
- Trained on an additional 500 billion tokens of mostly English web, math and code data
- Available on Kaggle, Hugging Face and Vertex AI Model Garden
- CodeGemma 1.1 followed on May 3, 2024
RecurrentGemma
RecurrentGemma is a research-oriented open Gemma model that replaces the transformer with recurrent layers and local attention (the Griffin architecture), lowering memory use and keeping sampling speed steady on long sequences.
- Modalities
- text → text
- Parameters
- 2B
- License
- Gemma Terms of Use
- At launch
- Open weights
- Higher throughput at large batch sizes than transformer-based Gemma
- Released on Kaggle with fine-tuning notebooks
- A 9B variant followed on June 11, 2024
text-embedding-004
text-embedding-004 is a Google text embedding model released in the Gemini API, offered in Vertex AI as text-embedding-preview-0409, with stronger retrieval performance than Google's earlier embedding models.
- Modalities
- text → embedding
- At launch
- API, Cloud platforms
- Supports elastic embedding sizes under 768 dimensions
- Released the same day Gemini 1.5 Pro entered public preview in the Gemini API
- Shut down in the Gemini API on Jan 14, 2026
Stable LM 2 12B
12B-parameter base and instruction-tuned language models trained on 2 trillion tokens in seven languages, released alongside an updated Stable LM 2 1.6B with tool use and function calling.
- Modalities
- text → text
- Parameters
- 12B
- License
- Stability AI Community License
- At launch
- Open weights
- Seven languages including English, Spanish, German and French
- Commercial use through a Stability AI Membership
- Base and chat models on Hugging Face
Command R+
Largest model in Cohere's R series at launch, built for advanced RAG with citations, multi-step tool use and multilingual business workloads.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 104B
- License
- CC-BY-NC-4.0
- At launch
- API, Open weights, Cloud platforms
- Launched first on Microsoft Azure and Cohere's API
- Covers 10 key business languages
- Research weights published under a non-commercial license
Stable Audio 2.0
Second-generation Stable Audio model producing full tracks with coherent structure up to three minutes long at 44.1 kHz stereo from a single prompt, and adding audio-to-audio generation.
- Modalities
- text, audio → audio
- At launch
- Consumer app, API
- Trained exclusively on licensed data from the AudioSparx library
- Free to use on the Stable Audio website at launch
- API availability followed later
Qwen1.5-32B
Qwen1.5-32B and Qwen1.5-32B-Chat are mid-size additions to the Qwen1.5 series, aimed at balancing performance, efficiency, and memory footprint at about 30 billion parameters.
- Modalities
- text → text
- Context
- 32K context
- Parameters
- 32B
- License
- Tongyi Qianwen License Agreement
- At launch
- Open weights
- Released as base and chat models
- Uses the same 32K context as other Qwen1.5 models
- Weights under the Tongyi Qianwen License Agreement
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 April 2024." The AI Engineering, updated Apr 30, 2024. https://theaiengineering.com/releases/2024/04
HTML link
<a href="https://theaiengineering.com/releases/2024/04">AI models released in April 2024</a> (The AI Engineering)- Full dataset
- 1,167 releases from 55 labs
- Coverage
- to
- Updated
- License
- CC BY 4.0
- Publisher
- The AI Engineering