ClovisLLM - GLM 5.1
Starting June 15, 2026, ClovisLLM will switch from GLM 4.7 to GLM 5.1. This migration aims to improve service performance across all advanced use cases: development, agents, long reasoning, tool use, and complex tasks.
No action is required on your part: the transition will be transparent for API users.
Why this evolution?
GLM 4.7 has been a solid foundation for ClovisLLM. However, our users' use cases are evolving: scenarios involving autonomous agents, advanced tool use, and long reasoning are becoming increasingly common. GLM 5.1 brings significant improvements on these axes, making this migration relevant for service quality.
This evolution is part of Clovis's continuous improvement approach: each version upgrade is an opportunity to deliver a more performant, more robust service that is better adapted to the real needs of our users.
What GLM 5.1 brings concretely
GLM 5.1 introduces notable advances across several dimensions:
- Long and scientific reasoning: improved ability to handle extended reasoning chains and problems with a strong logical or scientific component.
- Tool use and tool integration: significant gains in calling and orchestrating external tools (MCP, APIs, functions), with better understanding of the execution context.
- Agents and terminal tasks: sharply higher performance on autonomous agent scenarios, including command execution and navigation in complex environments.
- Code and security: marked progress on code generation and analysis tasks, including in cybersecurity contexts.
- Research and context management: improved ability to search, filter, and leverage information in extended contexts.
Performance comparison
The public indicators below illustrate the progression between GLM 4.7 and GLM 5.1 on several reference benchmarks. These figures reflect a generational evolution and serve as public indicators of progression.
Gray bars = GLM 4.7 · Colored bars = GLM 5.1
Key takeaways
The most notable progress concerns:
- CyberGym (+45.2 pts): the model's capability in code and security improves very significantly.
- Terminal-Bench 2.0 (+22.5 pts): agent scenarios in terminal environments benefit from a major performance leap.
- MCP-Atlas (+19.8 pts): tool usage via the MCP protocol is markedly improved.
- Tool-Decathlon (+16.9 pts) and BrowseComp (+16 pts): tool use and agent-based research progress substantially.
On scientific reasoning (GPQA-Diamond), GLM 5.1 confirms the already high level of GLM 4.7 with a marginal gain of +0.5 pt.
Positioning against market models
This comparison illustrates the highly competitive positioning of GLM 5.1 against the most advanced proprietary models on software development tasks. With an average score of 54.9 on SWE-Bench Pro, Terminal-Bench 2.0, and NL2Repo, GLM 5.1 approaches the best models on the market while retaining a decisive advantage for Clovis: its open-weight nature. This gives you a 100% sovereign model hosted in France by Claranet.
Impact for API users
The migration to GLM 5.1 is transparent: you do not need to make any changes to your existing integrations.
Concretely, here is what you can expect:
- Better output quality on development, long reasoning, and agent tasks.
- More reliable tool calls: functions, MCP, and external tools will be better orchestrated by the model.
- More relevant results on complex research tasks and navigation in extended contexts.
- Maintained compatibility: API endpoints, parameters, and response formats remain identical.
If you observe unexpected behavior after the migration, do not hesitate to contact us through the usual support channels.
Continuous improvement
This migration to GLM 5.1 is part of our ongoing work to deliver a performant LLM service adapted to real-world use cases. Each model evolution is guided by the concrete needs we observe: development, agents, tools, long reasoning, complex research, and API integration.
We will continue to evaluate new model generations to offer you the best possible performance, while maintaining API stability and compatibility.