Flash Launches Mark the Arrival of a Revolution in Agent Efficiency
Today marks the official release of Step3.7Flash. This open-source model is specifically designed to tackle the key challenges associated with the Agent era, including improving efficiency, ensuring reliability, and enhancing multimodal processing capabilities. It has quickly gained significant attention within the industry due to its release under an open weights model license (Apache 2.0).

Outstanding Benchmark Results and Strong Practical Performance
Step3.7Flash has delivered exceptional performance across a variety of critical evaluation tests:
First place in ClawEval-1.1 with 67.1 points
First place in SimpleVQA Search with 79.2 points
Second place in SWE-PRO with 56.3 points
A score of 95.3 in V* Python assessments
These results confirm its strong competitive edge in complex use cases such as Agent tasks, code generation, and visual search applications.
Optimized Parameters: Balancing Speed, Cost, and Performance
Developed specifically for Agentic, coding, searching, and multimodal workflows, Step3.7Flash has made significant advancements in speed and efficiency:
Reasoning Speed: Up to 400 TPS
Architecture: 198B sparse MoE structure featuring approximately 11B active parameters
Context Length: Supports up to 256K characters
Reasoning Level: Offers three different levels of reasoning depth
While maintaining high performance, this model also reduces actual deployment costs, making it an efficient choice for developers.
Multimodal Understanding Combined with Reliable Execution, Truly Enabling "See and Act"
The most notable feature of Step3.7Flash is its robust perception-action loop capability. It can interpret UI interfaces, charts, documents, and images, then automatically generate code or invoke tools to carry out the necessary actions.
In addition, its enhanced Web+ visual search function allows access to a wider range of information sources and supports more detailed follow-up queries. The reliability of tool calls has also been greatly improved, achieving a success rate of over 98% across all difficulty levels of the τ²-bench, which helps minimize issues such as target drift and tool call failures.
Compatibility with Existing Ecosystems and Easy Local Deployment
This model works well with popular agent frameworks including Claude Code, KiloCode, Hermes Agent, and OpenClaw, as well as MCP protocols. It is also compatible with local hardware such as Mac Studio M4Max, DGX Spark, and AMD AI Max+395, facilitating local deployment and use in scenarios where data privacy is a priority.
AIbase Commentary
Related article
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Swedish AI Startup Lovable Eyes $13.2 Billion Valuation After Major Funding Round
As AI-driven coding tools gain traction, Swedish startup Lovable has secured a major funding round. The company aims to raise $3 billion, potentially boosting its valuation to $13.2 billion—double the $6.6 billion recorded last December. Menlo Ventur
Google Tests Remy AI Agent for Gemini as Focus Shifts to User Control
According to Business Insider, Google is testing Remy, a new AI personal agent for Gemini. This tool aims to execute tasks on behalf of users, streamlining both professional workflows and daily routines.Currently, Remy is undergoing testing in an int
Related Special Topic Recommendations
Comments (0)
0/500
Today marks the official release of Step3.7Flash. This open-source model is specifically designed to tackle the key challenges associated with the Agent era, including improving efficiency, ensuring reliability, and enhancing multimodal processing capabilities. It has quickly gained significant attention within the industry due to its release under an open weights model license (Apache 2.0).

Outstanding Benchmark Results and Strong Practical Performance
Step3.7Flash has delivered exceptional performance across a variety of critical evaluation tests:
First place in ClawEval-1.1 with 67.1 points
First place in SimpleVQA Search with 79.2 points
Second place in SWE-PRO with 56.3 points
A score of 95.3 in V* Python assessments
These results confirm its strong competitive edge in complex use cases such as Agent tasks, code generation, and visual search applications.
Optimized Parameters: Balancing Speed, Cost, and Performance
Developed specifically for Agentic, coding, searching, and multimodal workflows, Step3.7Flash has made significant advancements in speed and efficiency:
Reasoning Speed: Up to 400 TPS
Architecture: 198B sparse MoE structure featuring approximately 11B active parameters
Context Length: Supports up to 256K characters
Reasoning Level: Offers three different levels of reasoning depth
While maintaining high performance, this model also reduces actual deployment costs, making it an efficient choice for developers.
Multimodal Understanding Combined with Reliable Execution, Truly Enabling "See and Act"
The most notable feature of Step3.7Flash is its robust perception-action loop capability. It can interpret UI interfaces, charts, documents, and images, then automatically generate code or invoke tools to carry out the necessary actions.
In addition, its enhanced Web+ visual search function allows access to a wider range of information sources and supports more detailed follow-up queries. The reliability of tool calls has also been greatly improved, achieving a success rate of over 98% across all difficulty levels of the τ²-bench, which helps minimize issues such as target drift and tool call failures.
Compatibility with Existing Ecosystems and Easy Local Deployment
This model works well with popular agent frameworks including Claude Code, KiloCode, Hermes Agent, and OpenClaw, as well as MCP protocols. It is also compatible with local hardware such as Mac Studio M4Max, DGX Spark, and AMD AI Max+395, facilitating local deployment and use in scenarios where data privacy is a priority.
AIbase Commentary
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Swedish AI Startup Lovable Eyes $13.2 Billion Valuation After Major Funding Round
As AI-driven coding tools gain traction, Swedish startup Lovable has secured a major funding round. The company aims to raise $3 billion, potentially boosting its valuation to $13.2 billion—double the $6.6 billion recorded last December. Menlo Ventur





Home






