Build Real-World Productivity Agents on SambaCloud with MiniMax 2.5

MiniMax 2.5 (M2.5) is the latest frontier model that is specialized for coding, agentic workflows, search, and real-world productivity. As a small Mixture of Experts (MoE) model, it delivers one of the best cost to intelligence ratios for these use cases, hitting 80.2% on SWE-Bench Verified, 51.3% on Multi-SWE-Bench, and 76.3% on BrowseComp (with full context management). It's 37% faster than its predecessor on complex agentic tasks, matching the throughput of Claude Opus 4.6 while costing a fraction of the price. SambaNova is running it fastest in the world—over 300 tokens/second.

Read the official MiniMax 2.5 announcement here.

MiniMax M2.5 is available now on the SambaCloud Enterprise Tier. Developer Tier access will be available next week as we expand capacity. Contact us today if you are interested in getting access through our enterprise plans.

MiniMax 2.5 is a great LLM for OpenClaw. Peter Steinberger (founder of OpenClaw) recommends MiniMax 2.5 over Claude.

Why Use MiniMax M2.5?

MiniMax M2.5 represents a massive leap forward from the M2.1 model, trained with RL in hundreds of thousands of real-world environments. Key upgrades include:

• SOTA Coding & Agentic Performance: Excels at full-stack development across 10+ languages (Go, Rust, Kotlin, and more), from zero-to-one system design to code review and testing.
• Search & Tool Use: Top-tier on BrowseComp and Wide Search, with better context management and precise search iterations.
• Office & Productivity: Built-in skills for Microsoft Word, PowerPoint, Excel, financial modeling, and domain-specific tasks like finance and law — ready for production agents.
• Efficiency Gains: 37% faster on SWE-Bench tasks, MoE architecture (230B total params, 10B active) for unmatched speed and cost (as low as $0.30/M input, $2.4/M output on Lightning). MiniMax M2.5 shines in non-reasoning mode for function calling and tool integration, making it a powerhouse for agentic frameworks like CrewAI, OpenClaw, or custom coding agents. Its efficient decomposition and token usage make it ideal for long-horizon tasks — perfect for coding or enterprise automation.
  
  

According to the Artificial Analysis Intelligence Index v4.0 (image courtesy of Artificial Analysis), MiniMax M2.5 ranks alongside top frontier models on applied benchmarks while landing firmly in the most attractive cost-to-intelligence quadrant. It delivers comparable real-world performance at significantly lower cost, making it one of the most efficient production models available today.

Get Started Quickly with SambaCloud

Powered by the SambaNova SN40L RDU dataflow architecture, SambaCloud delivers predictable high throughput, strong accuracy, and a developer-friendly API that scales from prototype to production. As an open-weight model, MiniMax M2.5 gives developers and enterprises full control to deploy privately and securely. SambaCloud is available in U.S. and Japan regions, giving organizations control over where their workloads and data are processed.

With just a few lines of Python, you can call MiniMax M2.5 to generate and refactor production-ready code, showcasing its ability to plan, optimize, and improve performance in a single pass.

Python

import os
from sambanova import SambaNova

# Initialize client using environment variable for security
client = SambaNova(
    api_key=os.environ["SAMBANOVA_API_KEY"],
    base_url="https://api.sambanova.ai/v1",
)

prompt = (
    "Write a Python script that fetches data from three APIs sequentially. "
    "Then refactor it to use asyncio for concurrent execution, add type hints, "
    "and explain the performance improvement."
)

response = client.chat.completions.create(
    model="MiniMax-M2.5",
    messages=[{"role": "user", "content": prompt}],
    temperature=0.1,  # low randomness for deterministic code output
    top_p=0.1
)

print(response.choices[0].message.content)

Check out SambaCloud and explore MiniMax M2.5 in the playground, or generate an API key and integrate it into your agentic application today.

MiniMax M2.5 brings applied intelligence to production. SambaNova delivers it at speed.