Salesforce logo

Salesforce models

Enterprise software company whose Salesforce AI Research group publishes open models such as BLIP-2, XGen, CodeT5+, CodeGen2.5, xGen-MM and the xLAM large action models.

San Francisco, United StatesFounded 1999salesforce.com

Releases
13
Open weights
12
Latest
Nov 06, 2026

November 2026

1 release
  1. MultimodalProprietary

    Koa Decisions

    Koa Decisions is Salesforce's decision model and the successor to HyperClassifier, which routes requests inside Agentforce. It reads a request and a set of user-defined options and returns one choice, a yes or no, or a score in a single step instead of generating text.

    Modalities
    text, image, audio → score
    • Built by applying Salesforce's own training techniques to Google's open-weight Gemma
    • Added image and audio input alongside text
    • Salesforce measured about 26 ms per Agentforce routing request, against about 1.4 seconds for a general-purpose LLM
    • At launch Salesforce offered a free Trailhead demo and said an API for developers and support in flows, Apex, and prompt templates would follow
    Source: Less Reasoning. More Doing. Meet Koa Decisions.

May 2025

1 release
  1. Small modelOpen weights

    xGen-small

    Compact enterprise-focused language models trained from scratch in 4B and 9B sizes, combining domain-focused data curation, length extension, instruction tuning and reinforcement learning for long-context performance at low cost.

    Modalities
    text → text
    Context
    125K context
    Parameters
    4B and 9B
    License
    CC-BY-NC-4.0
    At launch
    Open weights
    • Trained on a budget of eight trillion tokens
    • Base and instruct variants supported 128k-token context
    • Released for research use under CC BY-NC 4.0
    Source: xGen-small: Enterprise-ready Small Language Models

March 2025

1 release
  1. LLMOpen weights

    xLAM-2

    Second generation of Salesforce's large action models for function calling and AI agents, adding multi-turn conversation and tool use trained with the APIGen-MT data pipeline.

    Modalities
    text → text
    Parameters
    1B, 3B, 8B, 32B and 70B
    License
    CC-BY-NC-4.0
    At launch
    Open weights
    • Released in 1B, 3B, 8B, 32B and 70B sizes on Llama and Qwen backbones
    • Llama-based models supported 128k context; Qwen-based models 32k, extendable to 128k
    • Released for research use under CC BY-NC 4.0
    Source: SalesforceAIResearch/xLAM GitHub repository

January 2025

1 release
  1. EmbeddingOpen weights

    SFR-Embedding-Code

    Family of embedding models for code and text retrieval in 400M, 2B and 7B sizes, with the 400M and 2B models released openly.

    Modalities
    text → embedding
    Parameters
    400M and 2B
    License
    CC-BY-NC-4.0
    At launch
    Open weights
    • The 7B model ranked first on the CoIR code retrieval benchmark at announcement
    • The 400M model was designed for faster, lower-cost inference, including on CPUs
    • Open models released for research use under CC BY-NC 4.0
    Source: SFR-Embedding-Code: A Family of Embedding Models for Code Retrieval

September 2024

1 release
  1. LLMOpen weights

    xLAM

    Family of large action models optimized for function calling and AI agents, released for research in sizes for mobile use through high-performance deployments.

    Modalities
    text → text
    Parameters
    7B, 8x7B and 8x22B
    License
    CC-BY-NC-4.0
    At launch
    Open weights
    • Variants include xLAM-7b-r, xLAM-8x7b-r and xLAM-8x22b-r
    • Followed the July 17, 2024 release of the xLAM-1b-fc-r and xLAM-7b-fc-r function-calling models
    • Released for research under CC BY-NC 4.0
    Source: xLAM: A Family of Large Action Models for AI Agents - Salesforce

August 2024

1 release
  1. MultimodalOpen weights

    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
    Source: Salesforce/xgen-mm-phi3-mini-instruct-interleave-r-v1.5 · Hugging Face

January 2024

1 release
  1. EmbeddingOpen weights

    SFR-Embedding-Mistral

    Text embedding model from Salesforce AI Research built on E5-mistral-7b-instruct and Mistral-7B-v0.1, fine-tuned with multi-task training for retrieval, clustering and related tasks.

    Modalities
    text → embedding
    License
    CC-BY-NC-4.0
    At launch
    Open weights
    • Ranked first on the MTEB benchmark at release, averaging 67.6 across 56 datasets
    • Raised retrieval performance from 56.9 for E5-mistral-7b-instruct to 59.0
    • Released for research use under CC BY-NC 4.0
    Source: SFR-Embedding-Mistral: Enhance Text Retrieval with Transfer Learning

July 2023

1 release
  1. CodingOpen weights

    CodeGen2.5

    7B code generation model from Salesforce Research positioned to match larger 15B-plus code models at less than half the size, with local deployment in mind.

    Modalities
    text → text
    Parameters
    7B
    License
    Apache-2.0
    At launch
    Open weights
    • Successor to the CodeGen and CodeGen2 models
    • Released as multi, mono and instruct variants
    • Multi variant released under Apache 2.0
    Source: CodeGen2.5: Small, but mighty - Salesforce

June 2023

1 release
  1. LLMOpen weights

    XGen-7B

    Series of 7B language models trained with standard dense attention on sequences up to 8K tokens for up to 1.5 trillion tokens, plus instruction-tuned versions.

    Modalities
    text → text
    Context
    8K context
    Parameters
    7B
    License
    Apache-2.0
    At launch
    Open weights
    • Base models trained at 4K and 8K sequence lengths
    • Instruct versions tuned on public-domain instructional data
    • Base models released under Apache 2.0
    Source: Long Sequence Modeling with XGen: A 7B LLM Trained on 8K Input Sequence Length - Salesforce

May 2023

3 releases
  1. CodingOpen weights

    CodeT5+

    Family of open code language models with improved architectures and training, using a shallow encoder and deep decoder built on frozen code LLMs for efficient scaling.

    Modalities
    text → text
    Parameters
    220M to 16B
    License
    BSD-3-Clause
    At launch
    Open weights
    • Evaluated on more than 20 code benchmarks
    • Instruction-tuned InstructCodeT5+ 16B also released
    • Released in sizes from 220M to 16B
    Source: CodeT5+: Open Code Large Language Models
  2. MultimodalOpen weights

    InstructBLIP

    Vision-language instruction-tuning framework built on BLIP-2 models, released with state-of-the-art zero-shot generalization across a wide range of vision-language tasks.

    Modalities
    text, image → text
    At launch
    Open weights
    • Released in LAVIS with Vicuna 7B and 13B and FlanT5-XL and FlanT5-XXL variants
    • Vicuna variants used frozen Vicuna v1.1 weights that users prepared separately
    Source: salesforce/LAVIS GitHub repository: InstructBLIP
  3. CodingOpen weights

    CodeGen2

    Second generation of Salesforce's CodeGen program synthesis models, adding infill sampling and support for more programming languages than the original CodeGen family.

    Modalities
    text → text
    Parameters
    1B, 3.7B, 7B and 16B
    License
    Apache-2.0
    At launch
    Open weights
    • Released in 1B, 3.7B, 7B and 16B sizes
    • Introduced special tokens for infill sampling
    • Released under Apache 2.0
    Source: salesforce/CodeGen GitHub repository

January 2023

1 release
  1. MultimodalOpen weights

    BLIP-2

    Vision-language pre-training method and model family that connects frozen image encoders to frozen large language models through a lightweight Querying Transformer, enabling zero-shot instructed image-to-text generation.

    Modalities
    text, image → text
    At launch
    Open weights
    • Turned an 11B LLM into a multimodal model by training only 188M parameters
    • Released with OPT and FlanT5 language model variants
    • Code and models were open-sourced in Salesforce's LAVIS library and integrated into Hugging Face Transformers
    Source: BLIP-2: Scalable Multimodal Pre-training Method

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. "Salesforce models: release history and timeline." The AI Engineering, updated Nov 06, 2026. https://theaiengineering.com/releases/salesforce

HTML link

<a href="https://theaiengineering.com/releases/salesforce">Salesforce models: release history and timeline</a> (The AI Engineering)
Full dataset
1,167 releases from 55 labs
Coverage
to
Updated
License
CC BY 4.0
Publisher
The AI Engineering

Other labs