Model details
GLM-5.1 is positioned as Z.ai's next-generation flagship for agentic software engineering, succeeding GLM-5 with notably stronger coding abilities. The model's design centers on agentic workflows rather than single-shot answers: it breaks down ambiguous problems, runs experiments, reads results, and revises its strategy through repeated iteration. Independent provider coverage highlights its ability to work autonomously on complex tasks for up to eight hours, distinguishing it from predecessors that tend to plateau early in long sessions. Open weights are published under zai-org/GLM-5.1 on Hugging Face, alongside the associated technical report for the GLM-5 lineage and a public GitHub repository, signaling a fully open release rather than a closed-API-only rollout.
In benchmark terms, GLM-5.1 achieves state-of-the-art results on SWE-Bench Pro and leads GLM-5 by a wide margin on NL2Repo repository generation and Terminal-Bench 2.0, both of which probe real-world engineering depth rather than isolated coding snippets. Its longer-horizon behavior is the practical differentiator: the model sustains optimization across hundreds of rounds and thousands of tool calls, making it a strong fit for sustained engineering agents, repository-level code generation, and terminal-driven automation. Practical fit favors teams that need a deployable open-weights coding model for self-hosted pipelines or third-party inference such as SiliconFlow, especially where long-running autonomous workflows are central to the application.
Transparent token rates
Rates are shown per one million tokens. Combined means one million input plus one million output tokens.
SiliconFlow
In this article, we have tested and compared GLM-5.2 with Opus 4.8, GPT-5.5, and GLM-5.1 under the same conditions, highlighting where GLM-5.2 reaches frontier-level performance while offering a major cost advantage on SiliconFlow.
SiliconFlow
GLM-5.1, Z.AI's latest flagship model designed for long-horizon agentic engineering, achieves SOTA performance on SWE-Bench Pro and can work autonomously on complex tasks for up to 8 hours.
SiliconFlow
Compare DeepSeek-V3.1 and GLM-5.1 across performance, cost, capabilities, and real-world use cases. See which model fits your needs.
SiliconFlow
FriendliAI's changelog records that zai-org/GLM-5.1 was added to its Model APIs on April 7, 2026, matching the open-weights release date and confirming that the model was distributed to third-party inference providers in the same window as SiliconFlow's listing. This independent third-party availability signal is usefu Beyond GLM-5.1, the changelog tracks ongoing Z.ai model lifecycle activity on FriendliAI: zai-org/GLM-4.7 was deprecated on April 15, 2026, and the broader Model APIs lineup saw additions and deprecations across providers including Meta Llama, DeepSeek, Qwen, Cohere, and MiniMax in April through July 2026. The page als
SiliconFlow
Third-party benchmark aggregator LMSpeed lists GLM-5.1 as released in April 2026 with a reported context window of approximately 202.8K input tokens and 128K output tokens, and provides observed-capability estimates across coding (55.6, global rank 20), agents (54.7, global rank 19), reasoning (54.7) and math (52.1), w The LMSpeed entry points to OpenRouter, HuggingFace and Ollama as documentation/distribution channels rather than SiliconFlow specifically, so its figures should be read as provider-agnostic technical context for GLM-5.1 rather than SiliconFlow-specific pricing or quota data. Supported generation parameters include fre
SiliconFlow
SiliconFlow's official release notes confirm that on June 11, 2026, all traffic to the zai-org/GLM-5 model identifier was automatically routed to zai-org/GLM-5.1, with the successor offered at the same pricing as the original and developers urged to update their model names early to avoid future deprecation of GLM-5. T For developers, the practical implications are concrete: existing integrations specifying zai-org/GLM-5 will continue to work via automatic routing to GLM-5.1 without price changes, but model names should be updated proactively before GLM-5 is retired. The release notes do not provide benchmarks, context window, or mod
