Windsurf: GLM-5 and Minimax M2.5 Now Available
WindsurfWindsurf added two high-performance open-weight models to its IDE on February 17, 2026 — GLM-5 from Zhipu AI and Minimax M2.5. Both models are available to all users with limited-time promotional pricing and are integrated into Arena Mode's Frontier and Hybrid Arena battle groups. The additions bring Windsurf users competitive coding and reasoning capabilities at a fraction of the cost of leading proprietary models, with Minimax M2.5 scoring 80.2% on SWE-Bench Verified and GLM-5 claiming the top spot on the AA-Omniscience hallucination benchmark.
Sources & Mentions
5 external resources covering this update
MiniMax M2.5 released: 80.2% in SWE-bench Verified
Hacker News
GLM-5 is absolutely incredible for coding (7 new features)
Coding Beauty
Windsurf: GLM-5 and Minimax M2.5 are now available
X (Twitter)
Open source just pulled up to Opus 4.6 — at 1/20th the price
Hacker News
MiniMax-M2.5 vs GLM-5: Model Comparison
Artificial Analysis
GLM-5 and Minimax M2.5 Join Windsurf's Model Lineup
Windsurf expanded its AI model catalog on February 17, 2026, making two new frontier-class open-weight models available directly in the IDE: GLM-5 from Zhipu AI and Minimax M2.5. The additions represent a meaningful broadening of Windsurf's model diversity, giving developers access to capable, cost-efficient alternatives to proprietary models like Claude Opus 4.6 and Gemini 3 Pro.
Promotional Pricing
Both models launched with limited-time introductory credit pricing for all Windsurf users:
- GLM-5: 0.75x credits (limited-time)
- Minimax M2.5: 0.25x credits (limited-time)
At these rates, Minimax M2.5 in particular is positioned as an extremely cost-effective option, priced at roughly one-tenth to one-twentieth the credit cost of top proprietary models, while delivering competitive benchmark performance.
Arena Mode Integration
Both models are immediately available in Arena Mode's Frontier Arena and Hybrid Arena battle groups. This means Windsurf users can pit GLM-5 and Minimax M2.5 directly against other models in blind head-to-head sessions, providing a practical way to evaluate how these newcomers perform on real-world tasks before committing to them as primary assistants.
About GLM-5
GLM-5 is Zhipu AI's latest model and claims the top position on the AA-Omniscience hallucination evaluation benchmark — a 35-point improvement over its predecessor. Its standout characteristic is knowledge reliability: GLM-5 prioritizes factual accuracy and resists confabulation across science, code, and general knowledge domains. It positions itself close to Gemini 3 Pro in coding capability while offering stronger reasoning consistency.
About Minimax M2.5
Minimax M2.5 achieves 80.2% on SWE-Bench Verified, placing it among the highest-scoring publicly available models for software engineering tasks. The model is designed around agentic execution — before generating code, M2.5 actively decomposes tasks and plans structure from the perspective of a senior software architect. It leverages a highly sparse Mixture-of-Experts architecture to deliver token generation speeds exceeding 100–150 tokens per second on optimized infrastructure, making it well-suited for the iterative loops common in agentic coding workflows.
Bug Fixes in This Release
The February 17 release also included minor stability improvements: a fix for self-updating on Windows and a resolution for macOS UI flickering.