Z.ai icon

Z.ai GLM-5.1

GLM-5.1

Score 45.3 on Claude Code coding benchmark—94.6% of Claude Opus 4.6 performance—with 28% improvement over GLM-5, establishing a new frontier in cost-efficient agentic coding Generate code at 55+ tokens/sec with estimated 200K context window, enabling long-horizon multi-file refactoring and distributed system architecture design Access frontier-level coding intelligence from $3/month via Coding Plan with native compatibility for Claude Code, Cline, and Roo Code MCP tool integrations

Reviewed by ToolWorthy Editors·updated 3 months ago

Pricing:From $18/mo
Categories:
Newer version available·View latest
Jump to section

Featured alternatives

Coasty icon

Coasty

MakersClaw icon

MakersClaw

BrowserAct icon

BrowserAct

YAGNI icon

YAGNI

Timbal AI icon

Timbal AI

Kimi icon

Kimi

Pros & Cons

Pros

  • Scores 45.3 on Claude Code benchmarks—94.6% of Opus 4.6 at a fraction of the cost
  • 28% coding improvement over GLM-5 in a single generation step
  • Compatible with Claude Code, Cline, Kilo Code, and OpenCode without reconfiguration
  • Includes vision understanding and MCP integrations (web search, reader) built in
  • Official current Coding Plan docs start the Lite tier at $10/month, keeping GLM-5.1 comparatively accessible for solo developers.
  • Open-source release confirmed (MIT license, timeline TBD)

Cons

  • Standalone API not yet available; requires Coding Plan subscription
  • Context window (~200K) is smaller than Claude Opus's 1M for very long sessions
  • Benchmark evaluation was conducted in a Claude-optimized environment, so scores carry some bias
  • Open-source timeline is unconfirmed—no specific release date announced
  • Max plan pricing not publicly listed

Overview

GLM-5.1 is Z.AI's latest coding-focused model, released March 27, 2026, and available immediately to all GLM Coding Plan subscribers. Building on GLM-5-Turbo released just twelve days earlier, GLM-5.1 delivers a substantial coding performance leap—scoring 45.3 on Claude Code benchmarks versus the 35.4 baseline set by GLM-5. At 94.6% of Claude Opus 4.6's benchmark score, GLM-5.1 targets developers who want near-frontier coding intelligence at a materially lower list price than frontier subscriptions; current official Coding Plan docs start Lite at $10/month.

What's New

Coding Performance Leap

GLM-5.1's headline improvement is a 28% jump in coding benchmark scores compared to GLM-5. In evaluations run using Claude Code as the testing environment, GLM-5.1 achieved 45.3 points—2.6 points below Claude Opus 4.6's 47.9 score, or 94.6% of that result. The benchmark was run using Claude Code as the harness, so it is best treated as a Z.AI-reported comparative result rather than a neutral cross-vendor benchmark.

Public Z.AI materials position GLM-5.1 for complex coding and agentic workflows across a wide range of task types.

Generation Speed

GLM-5.1 maintains the 55+ tokens/sec generation speed introduced in the GLM-5 series, keeping interactive coding sessions fluid. Combined with an estimated 200K context window, developers can work across large, multi-file codebases without losing earlier context during long agent runs.

Vision & MCP Integrations

The GLM Coding Plan can be paired with a separate Vision MCP Server for screenshots, UI mockups, and diagrams. Z.AI's current docs describe that Vision MCP as powered by GLM-4.6V rather than as a new GLM-5.1-native capability. The GLM Coding Plan includes separate MCP services such as Web Search, Web Reader, and Zread. These are plan-level integrations documented across the Coding Plan docs, not newly introduced GLM-5.1 model-native features.

GLM-5.1 can be used in Claude Code, Cline, Kilo Code, OpenCode, OpenClaw, and other supported tools through Z.AI's Coding Plan integrations, although some tools require manual model configuration rather than true zero-config use.

Performance Benchmarks

Model Coding Score vs. Claude Opus 4.6
Claude Opus 4.6 47.9 100% (baseline)
GLM-5.1 45.3 94.6%
GLM-5-Turbo Official comparable score not disclosed on the current GLM-5.1 benchmark chart reviewed
GLM-5 35.4 73.9%

Evaluation conducted using Claude Code as the test harness. Claude-family models benefit from toolchain familiarity, so GLM-5.1's effective capability gap with Opus may be narrower than the 5.4-point difference indicates.

Generation speed: 55+ tokens/sec
Context window: 204,800 tokens
Max output: 131,072 tokens

Pricing & Plans

All GLM-5.1 access is through the GLM Coding Plan subscription. Standalone API access for GLM-5.1 is not yet available.

Plan Official current pricing/status 5-hour limit Weekly limit
Lite From $10/month Up to approx. 80 prompts Up to approx. 400 prompts
Pro From $30/month Up to approx. 400 prompts Up to approx. 2,000 prompts
Max Public docs reviewed list usage limits but do not clearly surface a sticker price in the docs cited here Up to approx. 1,600 prompts Up to approx. 8,000 prompts
  • All plans include GLM-5.1, GLM-5-Turbo, GLM-4.7, GLM-4.6, GLM-4.5, and GLM-4.5-Air; GLM-5 is currently listed for Pro and Max.
  • Requests reset within a rolling 5-hour window
  • No per-token charges within the subscription limit

Best For

  • Individual developers who want near-Opus coding quality at sub-$10/month
  • Teams running agentic coding workflows in Claude Code, Cline, or Kilo Code
  • Projects requiring vision-augmented coding (UI mockups, screenshot-driven debugging)
  • Developers building distributed systems or microservice architectures who need long-context reasoning
  • Cost-conscious teams evaluating Claude Opus alternatives for daily coding tasks
  • Developers who care about cost-effective coding-agent workflows in supported tools such as Claude Code, Cline, Kilo Code, and OpenClaw

FAQ

How does GLM-5.1 differ from GLM-5-Turbo?

GLM-5-Turbo (released March 15, 2026) focused on stability and efficiency for long-chain OpenClaw-style agent tasks. GLM-5.1 (released March 27, 2026) is shown on Z.AI's current benchmark chart at 45.3 versus Claude Opus 4.6 at 47.9; no official intermediate GLM-5-Turbo benchmark score has been published by Z.AI.

Can I use GLM-5.1 with Claude Code today?

Yes. GLM-5.1 is compatible with Claude Code via Z.AI's API compatibility layer. Configure the base URL and API key in your Claude Code settings to point to Z.AI's endpoint. The devpack documentation at docs.z.ai/devpack/using5.1 covers the exact configuration steps.

Is GLM-5.1 available as a standalone API?

As reviewed here, Z.AI documents GLM-5.1 primarily through the Coding Plan and supported-tool integrations. Standalone API access has not been announced; avoid assuming broader API availability until Z.AI publishes a dedicated model or API announcement.

When will GLM-5.1 be open-sourced?

Current public Z.AI materials do not provide a dated standalone GLM-5.1 repository link or confirmed MIT release timeline, so availability for local deployment remains unconfirmed.

What happens to request limits?

Requests are counted within a rolling 5-hour window: 120 for Lite, 600 for Pro. Limits reset continuously rather than on a fixed daily schedule, so consistent throughput is more achievable during active coding sessions.

Version History

GLM-5.2

Released on June 13, 2026

View Update
+What's new
3 updates
  • Use Z.ai's new flagship coding model across all GLM Coding Plan tiers, including Lite, Pro, Max, and Team plans, with 1M-context support for large codebases and long agent sessions
  • Configure Claude Code with `glm-5.2[1m]` plus `CLAUDE_CODE_AUTO_COMPACT_WINDOW=1000000`; configure OpenClaw/Cline-style tools with `glm-5.2`, 1M context settings, and 131K max output where the tool supports it.
  • Reserve GLM-5.2 for complex work because Z.ai treats it as a premium Opus-level model with higher quota multipliers, while GLM-4.7 remains the recommended option for routine tasks

GLM-5.1

Released on April 7, 2026

View Update
+What's new
3 updates
  • Use GLM-5.1 for long-horizon agentic engineering work, with stronger coding than GLM-5 and an official SWE-Bench Pro score of 58.4 against frontier coding models
  • Run multi-stage engineering tasks for up to 8 hours in one autonomous loop, covering planning, execution, testing, bug fixing, and production-grade delivery
  • Build agents that repeatedly experiment, analyze results, adjust strategy, and optimize systems, including benchmark-driven performance tuning workflows

GLM-5V-Turbo

Released on April 1, 2026

View Update
+What's new
3 updates
  • Process images, videos, design drafts, and document layouts natively as a multimodal vision coding model with 200K context window and 128K max output tokens for long-horizon agentic tasks
  • Execute the full perceive-plan-execute loop in GUI environments with leading scores on AndroidWorld, WebVoyager, and ZClawBench agent benchmarks optimized for OpenClaw workflows
  • Fuse visual understanding and code generation through CogViT vision encoder and 30+ task joint reinforcement learning across STEM, grounding, video, and coding domains

GLM-5.1

Current Version

Released on March 27, 2026

+What's new
3 updates
  • Score 45.3 on Claude Code coding benchmark—94.6% of Claude Opus 4.6 performance—with 28% improvement over GLM-5, establishing a new frontier in cost-efficient agentic coding
  • Generate code at 55+ tokens/sec with estimated 200K context window, enabling long-horizon multi-file refactoring and distributed system architecture design
  • Access frontier-level coding intelligence from $3/month via Coding Plan with native compatibility for Claude Code, Cline, and Roo Code MCP tool integrations

GLM-5-Turbo

Released on March 15, 2026

+What's new
3 updates
  • Execute complex OpenClaw agent workflows with superior tool invocation reliability, scheduled task continuity, and high-throughput long-chain execution optimized since training phase
  • Decompose and follow multi-layered complex instructions with enhanced comprehension, supporting collaborative task division among multiple agents and MCP tool integrations
  • Outperform GLM-5 across multiple ZClawBench task categories while supporting 200K context input with multiple thinking modes for dynamic, long-running agent tasks

GLM-5

Released on February 12, 2026

View Update
+What's new
3 updates
  • Handle complex systems engineering and long-horizon agentic tasks with 744B parameters (40B active) and DeepSeek Sparse Attention integration
  • Generate production-ready documents (.docx, .pdf, .xlsx) directly from text with built-in Agent mode and multi-turn collaboration
  • Execute code with best-in-class open-source performance on reasoning benchmarks, approaching frontier model capabilities

GLM-4.7-Flash

Released on January 19, 2026

+What's new
3 updates
  • Get lightweight version of GLM-4.7 with faster response times and high throughput optimized for real-time coding, writing, and translation tasks
  • Deploy efficiently with competitive performance at smaller scale while maintaining strong general capabilities across reasoning and content generation
  • Access free-tier model designed for high-frequency use cases with best-in-class aesthetic outputs, low latency, and simplified deployment

GLM-4.7-Flash

Released on January 19, 2026

+What's new
3 updates
  • Get lightweight version of GLM-4.7 with faster response times and high throughput optimized for real-time coding, writing, and translation tasks
  • Deploy efficiently with competitive performance at smaller scale while maintaining strong general capabilities across reasoning and content generation
  • Access free-tier model designed for high-frequency use cases with best-in-class aesthetic outputs, low latency, and simplified deployment

GLM-4.7

Released on December 22, 2025

+What's new
3 updates
  • Build cleaner modern webpages and professional slides with major improvements in UI aesthetics, visual quality, and accurate layout sizing for frontend development
  • Solve multilingual coding tasks faster with 73.8% on SWE-bench and 41% on Terminal Bench 2.0, delivering stronger performance across agent frameworks
  • Reason through complex mathematical and logical problems with 42.8% on HLE benchmark while enhancing tool-using and web browsing capabilities

GLM-4.6

Released on September 30, 2025

+What's new
3 updates
  • Handle longer conversations and complex multi-file codebases with expanded 200K context window, enabling more sophisticated agentic task execution
  • Code more efficiently in Claude Code, Cline, and Roo Code with superior benchmark performance and improved real-world coding accuracy
  • Leverage enhanced reasoning capabilities with native tool use support during inference, delivering stronger results in search-based agent workflows

GLM-4.5

Released on July 28, 2025

+What's new
3 updates
  • Unify reasoning, coding, and agentic capabilities in a single model delivering balanced performance across complex problem-solving and rapid content generation
  • Switch between thinking mode for deep analysis and non-thinking mode for instant responses, adapting intelligence level to task complexity on demand
  • Build full-stack web applications with stronger frontend quality, and integrate the model into Claude Code, Roo Code, or custom agent workflows through tool APIs

ChatGLM3-6B

Released on October 27, 2023

+What's new
3 updates
  • Execute code directly and invoke external tools with new Code Interpreter and Function Call capabilities, enabling autonomous agent-style task completion
  • Process information more accurately with improved training across 42 benchmarks covering semantics, mathematics, reasoning, code, and knowledge understanding
  • Deploy locally on consumer hardware with open-source 6B-parameter model supporting both academic research and free commercial use after registration

ChatGLM2-6B

Released on June 25, 2023

+What's new
3 updates
  • Handle longer conversations with expanded 32K context window using FlashAttention technology, enabling deeper multi-turn dialogue understanding
  • Get responses 42% faster with improved inference speed and INT4 quantization, supporting extended dialogues on consumer GPUs with only 6GB VRAM
  • Achieve stronger performance across reasoning and knowledge benchmarks with enhanced training on 1.4T bilingual tokens covering diverse domains

ChatGLM-6B

Released on March 14, 2023

+What's new
3 updates
  • Deploy locally on consumer-grade graphics cards with lightweight 6.2B-parameter bilingual model, enabling private ChatGPT-style conversations
  • Chat naturally in Chinese and English with open-source conversational AI trained on approximately 1 trillion tokens of diverse text data
  • Use freely for commercial purposes after simple registration, with full model weights and training code available for academic research

Top alternatives

Related categories

From the blog

View all →

Track Z.ai in ToolWorthy Weekly

Important tool updates, better alternatives, and selected AI signals in one weekly brief.

Weekly only. Unsubscribe anytime.