Average Ratings 0 Ratings

Total 0 out of 5 stars
ease 0 out of 5 stars
features 0 out of 5 stars
design 0 out of 5 stars
support 0 out of 5 stars

No User Reviews. Be the first to provide a review:

Write a Review

Average Ratings 0 Ratings

Total 0 out of 5 stars
ease 0 out of 5 stars
features 0 out of 5 stars
design 0 out of 5 stars
support 0 out of 5 stars

No User Reviews. Be the first to provide a review:

Write a Review

Description

Inkling 2 Mini is an open-weight AI reasoning and coding model designed to deliver advanced agentic capabilities while operating on comparatively modest computing hardware. Developed by Thinking Machines Lab, the model contains 26.4 billion parameters, of which 3.4 billion are active, and uses a specialized Mixture-of-Experts architecture to balance computational efficiency with performance. Its looped transformer design processes information through the model's middle layers three times per token, enabling additional internal reasoning without generating equivalent amounts of written reasoning. This architecture improves performance on complex software engineering tasks, long-context information retrieval, and multistep problem-solving activities. Inkling 2 Mini supports autonomous coding, tool calling, web research, document understanding, visual interpretation, and interactive application development. The model uses native NVFP4 and MXFP8 numerical precision to reduce memory requirements, allowing it to operate with approximately 20GB of model memory. Thinking Machines Lab reports that Inkling 2 Mini achieved 48.2% on DeepSWE v1.1 and 46.5% on Vals Index v2.2, outperforming other open-weight models under 100 billion parameters on those benchmarks at release. Developers can run the model locally through Ollama, llama.cpp, and other supported frameworks or access inference and fine-tuning capabilities through the Tinker platform. With downloadable model weights, reinforcement learning support, and integration with established AI development tools, Inkling 2 Mini enables organizations to build efficient AI applications without relying exclusively on hosted proprietary models.

Description

Step 3.5 Flash is a cutting-edge open-source foundational language model designed for advanced reasoning and agent-like capabilities, optimized for efficiency; it utilizes a sparse Mixture of Experts (MoE) architecture that activates only approximately 11 billion of its nearly 196 billion parameters per token, ensuring high-density intelligence and quick responsiveness. The model features a 3-way Multi-Token Prediction (MTP-3) mechanism that allows it to generate hundreds of tokens per second, facilitating complex multi-step reasoning and task execution while efficiently managing long contexts through a hybrid sliding window attention method that minimizes computational demands across extensive datasets or codebases. Its performance on reasoning, coding, and agentic tasks is formidable, often matching or surpassing that of much larger proprietary models, and it incorporates a scalable reinforcement learning system that enables continuous self-enhancement. Moreover, this innovative approach positions Step 3.5 Flash as a significant player in the field of AI language models, showcasing its potential to revolutionize various applications.

API Access

Has API Yes 

API Access

Has API Yes 

Screenshots View All

Screenshots View All

Integrations

GitHub No 
Hugging Face No 
Model Context Protocol (MCP) Yes 
ModelScope No 
Tinker Yes 
arXiv No 

Integrations

GitHub Yes 
Hugging Face Yes 
Model Context Protocol (MCP) No 
ModelScope Yes 
Tinker No 
arXiv Yes 

Pricing Details

$0.15 per 1M input tokens
$0.15 per 1M input tokens, $0.50 per 1M output tokens, and $0.015 per 1M cache hit tokens
Free Trial No 
Free Version No 

Pricing Details

Free
Free Trial No 
Free Version Yes 

Deployment

Web-Based Yes 
On-Premises No 
iPhone App No 
iPad App No 
Android App No 
Windows No 
Mac No 
Linux No 
Chromebook No 

Deployment

Web-Based Yes 
On-Premises No 
iPhone App Yes 
iPad App Yes 
Android App Yes 
Windows No 
Mac No 
Linux No 
Chromebook No 

Customer Support

Business Hours No 
Live Rep (24/7) No 
Online Support Yes 

Customer Support

Business Hours No 
Live Rep (24/7) No 
Online Support Yes 

Types of Training

Training Docs Yes 
Webinars No 
Live Training (Online) No 
In Person No 

Types of Training

Training Docs Yes 
Webinars No 
Live Training (Online) No 
In Person No 

Vendor Details

Company Name

Thinking Machines Lab

Founded

2025

Country

United States

Website

thinkingmachines.ai/

Vendor Details

Company Name

StepFun

Founded

2023

Country

China

Website

static.stepfun.com/blog/step-3.5-flash/

Alternatives

Alternatives

GPT-6 Luna Reviews

GPT-6 Luna

OpenAI
5.0 out of 5 stars
Inkling 2 Small Reviews

Inkling 2 Small

Thinking Machines Lab
MiMo-V2-Flash Reviews

MiMo-V2-Flash

Xiaomi Technology
Kimi K3.1 Pro Reviews

Kimi K3.1 Pro

Moonshot AI
5.0 out of 5 stars
Hy4 flash Reviews

Hy4 flash

Tencent
Inkling-Small Reviews

Inkling-Small

Thinking Machines Lab
5.0 out of 5 stars