AI models released in August 2024
24 releases from 17 labs in August 2024, including MiniMax Video-01, Command R and Command R+ (08-2024) and Qwen2-VL. Every entry is dated to the lab's official announcement and links to it.
- Releases
- 24
- Labs
- 17
- Open weights
- 16
August 2024
24 releasesMiniMax Video-01
Video-01 is MiniMax's first AI-native video generation model, producing 720p, 25 fps clips of up to six seconds with cinematic camera movement from a text prompt or a reference image plus text.
- Modalities
- text, image → video
- At launch
- API, Consumer app
- Supports text-to-video and image-to-video modes
- Became the model behind the Hailuo AI video product
- Available through the MiniMax API
Command R and Command R+ (08-2024)
Refreshed versions of Command R and Command R+ with better coding, math, reasoning and latency, plus Structured Outputs and selectable safety modes.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 32B and 104B
- License
- CC-BY-NC-4.0
- At launch
- API, Open weights, Cloud platforms
- Safety modes can be set to strict or contextual
- Available on Cohere's API, Amazon SageMaker and OCI
- Research weights released as c4ai-command-r-08-2024 and c4ai-command-r-plus-08-2024
Qwen2-VL
Qwen2-VL is Qwen's vision-language model series based on Qwen2, with improved understanding of images at various resolutions and videos longer than 20 minutes. Qwen open-sourced the 2B and 7B models and released an API for the 72B model.
- Modalities
- text, image, video → text
- Parameters
- 2B, 7B, 72B
- License
- Apache-2.0
- At launch
- API, Open weights
- Understands multilingual text inside images, including most European languages, Japanese, Korean, and Arabic
- 2B and 7B released under Apache 2.0
- Qwen2-VL-72B-Instruct weights followed on September 19, 2024
CogVideoX-5B
CogVideoX-5B is a larger open-weights model in Z.ai's CogVideoX text-to-video series, following CogVideoX-2B. Z.ai added an image-to-video variant, CogVideoX-5B-I2V, on September 19, 2024.
- Modalities
- text, image → video
- Parameters
- 5B
- License
- CogVideoX License
- At launch
- Open weights
- Larger sibling of CogVideoX-2B
- Image-to-video model CogVideoX-5B-I2V released September 19, 2024
- Weights under the CogVideoX License
Gemini 1.5 Flash-8B
Gemini 1.5 Flash-8B is a smaller and faster variant of Gemini 1.5 Flash, first released as an experimental model in the Gemini API and made production ready in October 2024 for high-volume and long-context tasks.
- Modalities
- text, image → text
- At launch
- API
- Experimental model gemini-1.5-flash-8b-exp-0827 released Aug 27, 2024, then updated as gemini-1.5-flash-8b-exp-0924
- Stable gemini-1.5-flash-8b-001 declared production ready on Oct 3, 2024
- Shut down in the Gemini API on Sep 29, 2025
Jamba 1.5 Mini and Jamba 1.5 Large
Open SSM-Transformer model family with a 256K effective context window, offered for download and through AI21 Studio and major cloud platforms.
- Modalities
- text → text
- Context
- 250K context
- Parameters
- MoE: Mini 52B total, 12B active; Large 398B total, 94B active
- License
- Jamba Open Model License
- At launch
- Open weights, API, Cloud platforms
- Mini handles up to 140K tokens on a single GPU
- Available on Google Cloud Vertex AI, Microsoft Azure and NVIDIA NIM at launch
- Jamba Open Model License permits research and commercial use
Phi-3.5-mini
Phi-3.5-mini is a 3.8B open Microsoft model that updates Phi-3-mini with further multilingual pre-training and post-training, improving multilingual, multi-turn conversation and reasoning quality while supporting a 128K context.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- 3.8B
- License
- MIT
- At launch
- Open weights, API
- 3.8B dense model with a 128K token context length
- Lists support for 23 languages
- Offered as a serverless endpoint in Azure AI Studio
Phi-3.5-MoE
Phi-3.5-MoE is Microsoft's first mixture-of-experts Phi model, activating two of 16 experts per token for 6.6B active parameters, with multilingual support and a 128K context length.
- Modalities
- text → text
- Context
- 125K context
- Parameters
- MoE: 16x3.8B experts, 6.6B active
- License
- MIT
- At launch
- Open weights
- Mixture-of-experts with 16 experts and 6.6B active parameters
- 128K token context length
- Supports more than 20 languages
Phi-3.5-vision
Phi-3.5-vision is a 4.2B open multimodal Microsoft model that adds multi-frame image understanding to the Phi-3 vision line, enabling image comparison, multi-image summarization and video summarization.
- Modalities
- text, image → text
- Context
- 125K context
- Parameters
- 4.2B
- License
- MIT
- At launch
- Open weights
- 4.2B parameters with a 128K token context length
- Adds multi-frame image understanding and reasoning
- Not optimized for multilingual use cases
Ideogram 2.0
Ideogram text-to-image model released free to all users, launched together with the Ideogram iOS app, a beta Ideogram API and Ideogram Search.
- Modalities
- text → image
- At launch
- Consumer app, API
- Five styles: General, Realistic, Design, 3D and Anime
- Color palette control, including custom palettes
- Design style renders longer, more accurate text
Mistral-NeMo-Minitron 8B
Small language model created by pruning and distilling the 12B Mistral NeMo model down to 8B parameters while recovering accuracy through retraining.
- Modalities
- text → text
- Parameters
- 8B
- License
- NVIDIA Open Model License
- At launch
- Open weights, API
- Derived from Mistral NeMo 12B
- Offered as an NVIDIA NIM microservice and on Hugging Face
- Trained between July 24 and August 10, 2024
xGen-MM v1.5 (BLIP-3)
Series of open multimodal models from Salesforce's xGen-MM family, also known as BLIP-3, including a model that handles interleaved images and text.
- Modalities
- text, image → text
- License
- Apache-2.0
- At launch
- Open weights
- Built on a Phi-3-mini language backbone
- Follows the first xGen-MM (v1) release
- Interleave instruct model released under Apache 2.0
DeepSeek-Prover-V1.5
DeepSeek-Prover-V1.5 is a 7B open-weights model for theorem proving in Lean 4, pre-trained on DeepSeekMath-Base and refined with reinforcement learning from proof assistant feedback. It improves on DeepSeek-Prover-V1, whose weights were also published.
- Modalities
- text → text
- Parameters
- 7B
- License
- DeepSeek Model License
- At launch
- Open weights
- Released as Base, SFT, and RL models
- Uses reinforcement learning from proof assistant feedback (RLPAF)
- Model use subject to the DeepSeek model license, which supports commercial use
Hermes 3
Steerable instruction-tuned models created by fine-tuning Llama 3.1 8B, 70B and 405B on primarily synthetic data, with long-term context retention and multi-turn chat.
- Modalities
- text → text
- Parameters
- 8B, 70B and 405B
- License
- Llama 3 Community License
- At launch
- Open weights
- Released with the Hermes 3 technical report
- Supports complex roleplaying and internal monologue
- A Llama 3.2 3B version followed in December 2024
Gen-3 Alpha Turbo
Faster, lower-cost version of Gen-3 Alpha that generates seven times faster while matching the original's performance across many use cases.
- Modalities
- text, image → video
- At launch
- Consumer app, API
- Seven times faster than Gen-3 Alpha
- Gained keyframes, extensions, video to video and camera control in later updates
- Integrated into apps through the Runway API
Grok-2 and Grok-2 mini
Grok-2 is xAI's successor to Grok-1.5 with improved chat, coding, and reasoning, released in beta for X Premium and Premium+ users alongside Grok-2 mini, a smaller sibling balancing speed and answer quality.
- Modalities
- text, image → text
- At launch
- Consumer app
- An early version was tested on LMSYS Chatbot Arena as sus-column-r
- Grok-2 supports both text and vision understanding
- xAI experimented with Black Forest Labs' FLUX.1 for image generation on X
Qwen2-Audio
Qwen2-Audio is the next version of Qwen-Audio, accepting audio and text input for voice chat and audio analysis. It understands more than 8 languages and dialects.
- Modalities
- text, audio → text
- Parameters
- 7B
- License
- Apache-2.0
- At launch
- Open weights
- Released as Qwen2-Audio-7B and Qwen2-Audio-7B-Instruct
- Supports voice chat and audio analysis modes
- Weights on Hugging Face and ModelScope under the Apache 2.0 license
Qwen2-Math
Qwen2-Math is Qwen's first series of math-specific language models, initialized from Qwen2 1.5B, 7B, and 72B and pre-trained on a mathematics-specific corpus. It mainly supports English at launch.
- Modalities
- text → text
- Parameters
- 1.5B, 7B, 72B
- License
- Apache-2.0 (72B uses the Tongyi Qianwen License Agreement)
- At launch
- Open weights
- Released as base and instruct models
- Bilingual English and Chinese math models were announced as coming
- Smaller sizes under Apache 2.0
Amazon Titan Image Generator v2
Amazon Titan Image Generator v2 updates Amazon's image model with image conditioning from reference images, color palette guidance, background removal and subject consistency through fine-tuning.
- Modalities
- text, image → image
- At launch
- API
- Adds image conditioning with Canny edge or segmentation control
- Adds color palette guidance and background removal
- Generally available in Amazon Bedrock in US East (N. Virginia) and US West (Oregon)
CogVideoX-2B
CogVideoX-2B is the first open-weights model in Z.ai's CogVideoX text-to-video series, released with the 3D Causal VAE it uses to reconstruct video. It is a smaller entry-level video generation model.
- Modalities
- text → video
- Parameters
- 2B
- License
- Apache-2.0
- At launch
- Open weights
- 3D Causal VAE open-sourced alongside the model
- Weights on Hugging Face under the Apache 2.0 license
- Followed by CogVideoX-5B on August 27, 2024
gpt-4o-2024-08-06
gpt-4o-2024-08-06 is a GPT-4o snapshot trained to follow complex JSON schemas, released with Structured Outputs in the API, which constrains output to developer-supplied JSON Schemas via strict function calling or response_format.
- Modalities
- text, image → text
- Context
- 125K context
- At launch
- API
- response_format json_schema works with gpt-4o-2024-08-06 and gpt-4o-mini-2024-07-18
- The gpt-4o alias points to this snapshot per the docs model page
- 128,000-token context and 16,384 max output tokens per the gpt-4o docs page
- Fine-tuning for this snapshot became generally available on Aug 20, 2024
FLUX.1 [dev]
Open-weight, guidance-distilled FLUX.1 model distilled from FLUX.1 [pro], released for non-commercial use with similar quality and prompt adherence while being more efficient than a standard model of its size.
- Modalities
- text → image
- Parameters
- 12B
- License
- FLUX.1 [dev] Non-Commercial License
- At launch
- Open weights, API, Cloud platforms
- Weights on Hugging Face under a non-commercial license
- Commercial use available through BFL licensing
- 12B parameters, same architecture as the FLUX.1 suite
FLUX.1 [pro]
Flagship model of the FLUX.1 text-to-image suite that launched alongside Black Forest Labs itself, built on a 12B-parameter hybrid transformer flow-matching architecture and offered through the BFL API and partners.
- Modalities
- text → image
- Parameters
- 12B
- At launch
- API, Cloud platforms
- Launched with the FLUX.1 [dev] and [schnell] variants
- Hybrid multimodal and parallel diffusion transformer blocks scaled to 12B parameters
- Available via the BFL API, Replicate and fal.ai
FLUX.1 [schnell]
Fastest FLUX.1 variant, a few-step text-to-image model tailored for local development and personal use and released openly under the Apache 2.0 license.
- Modalities
- text → image
- Parameters
- 12B
- License
- Apache-2.0
- At launch
- Open weights, Cloud platforms
- Apache 2.0 license
- Inference code on GitHub and in Hugging Face Diffusers
- Day-one ComfyUI integration
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 August 2024." The AI Engineering, updated Aug 31, 2024. https://theaiengineering.com/releases/2024/08
HTML link
<a href="https://theaiengineering.com/releases/2024/08">AI models released in August 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