AI models released in July 2024
20 releases from 12 labs in July 2024, including Gemma 2 2B, ShieldGemma and Midjourney V6.1. Every entry is dated to the lab's official announcement and links to it.
- Releases
- 20
- Labs
- 12
- Open weights
- 16
July 2024
20 releasesGemma 2 2B
Gemma 2 2B is a new 2B-parameter version of Gemma built on the Gemma 2 architecture, designed to run efficiently on hardware from phones and laptops to cloud deployments on Vertex AI and GKE.
- Modalities
- text → text
- Parameters
- 2B
- License
- Gemma Terms of Use
- At launch
- Open weights, API, Cloud platforms
- Weights on Kaggle, Hugging Face and Vertex AI Model Garden
- Can be tried in Google AI Studio
- Released under Gemma terms for research and commercial use
ShieldGemma
ShieldGemma is a set of safety classifiers built on Gemma 2 that filter the inputs and outputs of AI models. It detects hate speech, harassment, sexually explicit content and dangerous content.
- Modalities
- text → text
- Parameters
- 2B, 9B, 27B
- License
- Gemma Terms of Use
- At launch
- Open weights
- 2B suits online classification; 9B and 27B target offline use
- Released alongside Gemma 2 2B and the Gemma Scope interpretability tools
Midjourney V6.1
Point release of V6 with more coherent bodies and objects, better image quality and small details, new 2x upscalers, improved text accuracy and a new personalization model, made default at launch.
- Modalities
- text, image → image
- At launch
- Consumer app
- About 25% faster for standard image jobs
- Default model until June 16, 2025
- Personalization code versioning
SAM 2
SAM 2 is Meta's next-generation Segment Anything Model, a unified model for real-time promptable object segmentation in both images and videos.
- Modalities
- image, video → image, video
- License
- Apache-2.0
- At launch
- Open weights
- Runs at about 44 frames per second
- Released with the SA-V dataset of about 51,000 videos under CC BY 4.0
- Treats an image as a single-frame video so one architecture handles both
DeepSeek-Coder-V2-0724
DeepSeek-Coder-V2-0724 is an updated DeepSeek-Coder-V2 checkpoint with improved general capabilities through alignment optimization. It upgraded the deepseek-coder API model and was later merged into DeepSeek-V2.5.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- MoE: 236B total, 21B active
- License
- DeepSeek Model License
- At launch
- API, Open weights
- Upgraded the deepseek-coder API model on July 24, 2024
- Weights published on Hugging Face as DeepSeek-Coder-V2-Instruct-0724
- One of the two models merged to create DeepSeek-V2.5
Mistral Large 2
Second-generation flagship with 123B parameters and a 128K-token context window, designed for single-node inference and released with weights under the Mistral Research License.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 123B
- License
- Mistral Research License
- At launch
- Open weights, API
- Supports dozens of languages and 80+ coding languages
- Commercial self-deployment requires a Mistral Commercial License
- API model ID mistral-large-2407
Llama 3.1
Llama 3.1 introduced the 405B model, which Meta described as the largest openly available foundation model, plus upgraded multilingual 8B and 70B models, all with a 128K context window.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 8B, 70B, 405B
- License
- Llama 3.1 Community License
- At launch
- Open weights, Cloud platforms
- Supports eight languages
- License changed to allow using Llama outputs to improve other models
- 405B trained on over 15 trillion tokens
- Over 25 partners, including AWS, Azure and Google Cloud, offered services on day one
Llama Guard 3
Llama Guard 3 is a multilingual content safety classifier fine-tuned from Llama 3.1 8B, released with Llama 3.1 and optimized for search and code interpreter tool calls.
- Modalities
- text → text
- Parameters
- 8B
- License
- Llama 3.1 Community License
- At launch
- Open weights
- Predicts safety labels across 14 categories, adding Code Interpreter Abuse
- Supports eight languages
Prompt Guard
Prompt Guard is a small classifier released with Llama 3.1 that labels inputs as benign, prompt injection or jailbreak, built to filter prompt attacks in LLM applications.
- Modalities
- text → text
- Context
- 1K context
- Parameters
- 86M
- License
- Llama 3.1 Community License
- At launch
- Open weights
- Fine-tuned from mDeBERTa-v3-base with 86M backbone parameters
- 512-token context window; longer inputs are split into segments
Rerank 3 Nimble
Faster member of the Rerank 3 series, about three times faster than Rerank 3 while keeping high accuracy, offered in English and multilingual versions.
- Modalities
- text → score
- At launch
- API
- Multilingual version covers 100+ languages
- Can rerank documents tens of thousands of tokens long
- Handles emails, JSON and tabular data
Eleven Turbo v2.5
Low-latency multilingual model that made Hindi, French, Spanish, Mandarin and 27 other languages three times faster and English 25% faster for conversational AI.
- Modalities
- text → audio
- At launch
- API, Consumer app
- First support for Vietnamese, Hungarian and Norwegian
- Three times faster in non-English languages
- Later outclassed by Flash v2.5
GPT-4o mini
GPT-4o mini is OpenAI's cost-efficient small model, supporting text and vision in the API and replacing GPT-3.5 in ChatGPT for Free, Plus, and Team users. It is the first API model to apply OpenAI's instruction hierarchy method.
- Modalities
- text, image → text
- Context
- 125K context
- At launch
- API, Consumer app
- API model ID gpt-4o-mini-2024-07-18
- 128K context window, up to 16K output tokens, knowledge up to October 2023
- Available in the Assistants, Chat Completions, and Batch APIs; fine-tuning launched Jul 23, 2024
- ChatGPT Enterprise access began the following week
Granite Code 128K (3B, 8B)
Long-context versions of the 3B and 8B Granite Code models that extend the context window to 128K tokens, released as base and instruct checkpoints.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 3B and 8B
- License
- Apache-2.0
- At launch
- Open weights
- Extends Granite Code context to 128K tokens
- Base and instruct checkpoints for each size
- Released under Apache 2.0
Mistral NeMo
12B model built with NVIDIA that offers a 128K-token context window, multilingual skills and a drop-in replacement path for systems using Mistral 7B.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 12B
- License
- Apache-2.0
- At launch
- Open weights, API
- Introduced the Tekken tokenizer, trained on more than 100 languages
- Trained with quantization awareness for FP8 inference
- Base and instruct checkpoints under Apache 2.0
Codestral Mamba
Mamba-architecture code model designed with help from Albert Gu and Tri Dao, offering linear-time inference and tested on in-context retrieval up to 256K tokens.
- Modalities
- text → text
- Context
- 250K context
- Parameters
- 7.3B
- License
- Apache-2.0
- At launch
- Open weights, API
- Instructed model with 7,285,403,648 parameters
- Released under Apache 2.0
- API model ID codestral-mamba-2407
Mathstral 7B
STEM-specialized model built on Mistral 7B for multi-step mathematical reasoning, produced in collaboration with Project Numina and released with open weights.
- Modalities
- text → text
- Context
- 32K context
- Parameters
- 7B
- License
- Apache-2.0
- At launch
- Open weights
- Instructed model for math and STEM subjects
- Weights hosted on Hugging Face
- Can be fine-tuned with mistral-finetune
SmolLM
Family of small language models trained on SmolLM-Corpus, a curated dataset that combines synthetic textbooks, educational Python code and educational web data.
- Modalities
- text → text
- Parameters
- 135M, 360M and 1.7B
- License
- Apache-2.0
- At launch
- Open weights
- Released in 135M, 360M and 1.7B sizes
- Training corpus includes Cosmopedia v2 and Python-Edu
- Released under Apache 2.0
Granite 20B FunctionCalling
Function calling model fine-tuned from Granite 20B Code Instruct with a multi-task training approach to bring tool-calling abilities to the Granite family.
- Modalities
- text → text
- Parameters
- 20B
- License
- Apache-2.0
- At launch
- Open weights
- Based on granite-20b-code-instruct
- Trained with a multi-task function calling approach
- Released under Apache 2.0
Kolors
Kolors is Kuaishou's large text-to-image model trained on billions of text-image pairs, bilingual in Chinese and English, open-sourced at the 2024 World Artificial Intelligence Conference.
- Modalities
- text → image
- License
- Kolors Model License Agreement
- At launch
- Open weights
- Bilingual Chinese and English prompts with a 256-token text context
- ControlNet, inpainting, and IP-Adapter variants released in July 2024
- Weights free for academic research; commercial use requires registration
CodeGeeX4-ALL-9B
CodeGeeX4-ALL-9B is the open-weights version of Z.ai's CodeGeeX4 code model series, continually trained on GLM-4-9B. A single model supports code completion and generation, code interpreter, web search, function calling, and repository-level code Q&A.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 9B
- License
- CodeGeeX4 License
- At launch
- Open weights
- 128K context length
- Available on Hugging Face, ModelScope, WiseModel, and Ollama
- Weights under the CodeGeeX4 License
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 July 2024." The AI Engineering, updated Jul 31, 2024. https://theaiengineering.com/releases/2024/07
HTML link
<a href="https://theaiengineering.com/releases/2024/07">AI models released in July 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