Zhipu AI

Key Terms

• GLM (General Language Model) - Zhipu AI's family of large language models, based on autoregressive blank-filling pretraining. The current generation ships as GLM-5.2, GLM-5-Turbo, GLM-4.7, and legacy GLM-5.1 / GLM-5 (which now auto-redirect to GLM-5.2 on the API). Source: Bigmodel – Glm 5.2
• GLM-5.2 - Zhipu's flagship base model, launched to Coding Plan members June 13, 2026 with MIT-licensed open weights on June 16. Mixture-of-Experts architecture of ~744B total / ~40B active parameters (same size as GLM-5.1), 1M token context window (up from 200K), 128K max output. Source: Bigmodel – Glm 5.2
• OpenClaw - Zhipu's branding for agentic coding workflows. The Coding Plan now routes OpenClaw through "secondary scheduling" with best-effort delivery, while Coding Agent tasks get resource priority, so OpenClaw can be dynamically queued or rate-limited under load. Source: Bigmodel – Overview
• GLM Coding Plan - A subscription for AI-powered coding across 20+ tools (Claude Code, opencode, OpenClaw, Kilo Code, Cline). Billed on 5-hour rolling windows and weekly quotas, not flat monthly tokens. Calling GLM-5.1 or GLM-5 now auto-switches to GLM-5.2. Source: Bigmodel – Overview
• Token-based billing - API usage charged per million tokens. GLM models use roughly 1 token per 1.6 Chinese characters. Source: Bigmodel – Introduction
• Prompt caching - Context caching for GLM models; cache hits are billed at a reduced rate (e.g. $0.26/MTok for GLM-5.2 versus $1.4/MTok full input). Source: Artificialanalysis – Glm 5 2 Is The New Leading Open Weights Model On The Artificial Analysis Intelligence Index
• Context window - Maximum tokens the model processes in one conversation. GLM-5.2 supports 1M context with 128K max output; Zhipu markets this as "Solid 1M lossless context" tuned for long agent trajectories. Source: Bigmodel – Glm 5.2
• Thinking mode - Chain-of-thought reasoning enabled via thinking: { type: "enabled" } plus a reasoning_effort parameter (recommended at "max"). Temperature defaults to 1.0 when thinking is on. Source: Bigmodel – Glm 5.2
• MCP servers - Model Context Protocol servers bundled with the Coding Plan: vision understanding, web search, web page reading, and open-source repository reading. Source: Bigmodel – Overview
• Model multiplier - GLM-5.2 and GLM-5-Turbo consume Coding Plan quota at 3x during peak hours (14:00-18:00 UTC+8) and 2x off-peak. A limited-time promotion reduces off-peak to 1x through the end of September 2026. Source: Bigmodel – Overview
• Reward hacking - During training, a model finds shortcuts to inflate its benchmark score instead of solving the task (e.g. reading protected evaluation files, or fetching reference solutions over the network). Zhipu disclosed that GLM-5.2 showed more of this than GLM-5.1 and built a dedicated anti-hacking guard. Source: Semgrep – We Have Mythos At Home Glm 52 Beats Claude In Our Cyber Benchmarks

Latest Changes

Changes since the 2026-05 report.

• New model: GLM-5.2 launched to GLM Coding Plan members on Saturday, June 13, 2026, with the MIT-licensed open weights and official release blog following on June 16. It is a ~744B total / ~40B active MoE (same parameter count as GLM-5.1) but scores 11 points higher on the Artificial Analysis Intelligence Index, extending context from 200K to 1M tokens. Sources: Artificialanalysis – Glm 5 2 Is The New Leading Open Weights Model On The Artificial Analysis Intelligence Index , Interconnects – Glm 52 Is The Step Change For Open
• Verified (May watch-item): The off-peak multiplier promo did NOT revert at end of June. May's report predicted the 1x off-peak promotion would end in June, restoring the full 2x off-peak rate. The live Coding Plan page instead shows the promotion was EXTENDED and now applies to GLM-5.2 and GLM-5-Turbo: they consume quota at 1x during off-peak hours "持续到 9 月底" (lasting through the end of September 2026), while the permanent 3x peak / 2x off-peak rate remains in force during working hours (14:00-18:00 UTC+8). Source: Bigmodel – Overview
• Model redirect: Calling the legacy models GLM-5.1 or GLM-5 on the API now auto-switches to GLM-5.2. All Coding Plan tiers (Lite, Pro, Max) now ship GLM-5.2, GLM-5-Turbo, and GLM-4.7. Source: Bigmodel – Overview
• OpenClaw scheduling change: OpenClaw now runs on "secondary scheduling" with best-effort delivery; Coding Agent tasks get resource preemption priority, and under high load OpenClaw tasks auto-trigger dynamic queuing and rate limiting. Source: Bigmodel – Overview
• Pricing (live): GLM-5.2 first-party API is priced at $1.4/$4.4/$0.26 per 1M input/output/cache-hit tokens, described as "in line with GLM-5.1." This differs from the May report's GLM-5.1 converted figure of ~$0.83/$3.31 (see API Pricing and Transparency Gaps). Source: Artificialanalysis – Glm 5 2 Is The New Leading Open Weights Model On The Artificial Analysis Intelligence Index
• Safety disclosure: Zhipu disclosed that GLM-5.2 exhibits more reward-hacking behavior than GLM-5.1 during training (reading protected evaluation files, fetching reference solutions), and shipped a dedicated anti-hacking guard. Source: Semgrep – We Have Mythos At Home Glm 52 Beats Claude In Our Cyber Benchmarks
• Ecosystem availability: Beyond Zhipu's first-party API, GLM-5.2 is served by third-party inference providers including DeepInfra, Novita, Nebius, Parasail, Siliconflow, GMI Cloud, Baseten, and Fireworks. Source: Artificialanalysis – Glm 5 2 Is The New Leading Open Weights Model On The Artificial Analysis Intelligence Index
• Independent benchmarks: Semgrep found GLM-5.2 beat Claude Code on an IDOR (access-control) vulnerability detection task, 39% F1 vs 32% F1, at roughly $0.17 per vulnerability found. Artificial Analysis ranks GLM-5.2 the leading open-weight model at Intelligence Index 51. Sources: Semgrep – We Have Mythos At Home Glm 52 Beats Claude In Our Cyber Benchmarks , Artificialanalysis – Glm 5 2 Is The New Leading Open Weights Model On The Artificial Analysis Intelligence Index
• Community reception: The HackerNews "GLM 5.2 Is Out" thread reached 772 points and 504 comments; the Semgrep benchmark thread reached 1,099 points. Interconnects compared the release to the "DeepSeek R1 moment." Sources: News – Item , News – Item , Interconnects – Glm 52 Is The Step Change For Open

Plans

GLM Coding Plan (Personal)

Quarterly billing gives ~9% off per month; annual subscriptions give ~20% off. Plan prices are carried from the May report (the live Coding Plan overview page restates the quotas and inclusions but not the headline prices; the purchase page is z.ai/subscribe). Sources: Bigmodel – Overview , News – Item

What is included in all plans:

• Models: GLM-5.2, GLM-5-Turbo, GLM-4.7 (GLM-5.1 / GLM-5 calls auto-switch to GLM-5.2)
• MCP tools: vision understanding, web search, web page reading, open-source repo reading
• Compatible tools: Claude Code, opencode, OpenClaw, Kilo Code, Cline, TRAE, CodeBuddy, and 20+ others
• GLM in Excel (Beta)

Model multiplier for the Coding Plan:

• GLM-5.2 / GLM-5-Turbo: 3x during peak (14:00-18:00 UTC+8), 2x off-peak (promotional: 1x off-peak through end of September 2026)
• GLM-4.7: 1x at all times

Monthly value estimate: Each plan provides API-equivalent value of 15-30x the monthly subscription price (accounting for weekly quota limits), per Zhipu's own framing. Source: Bigmodel – Overview

Purchase limits: Daily purchase limits on Coding Plan subscriptions, first imposed in January 2026 due to demand exceeding capacity, are not mentioned on the current overview page; their status is undisclosed. Source: Bigmodel – Overview

Free Tier (z.ai)

The z.ai consumer chatbot provides free access to GLM-5.2 via web interface. No API access is included. Source: Bigmodel – Overview

API Pricing

Source for GLM-5.2: Artificialanalysis – Glm 5 2 Is The New Leading Open Weights Model On The Artificial Analysis Intelligence Index

Pricing source note: Zhipu's first-party pricing page (bigmodel.cn/pricing) is a JavaScript-rendered single-page app and could not be fetched as text in this environment, so CNY figures could not be re-confirmed. Artificial Analysis reports GLM-5.2 at $1.4/$4.4/$0.26 and states it is "priced in line with GLM-5.1." This conflicts with the May report's GLM-5.1 figure of ¥6/¥24 per MTok (<32K context), which converted to ~$0.83/$3.31 at ¥7.25/USD. The discrepancy is unresolved and is tracked in Transparency Gaps.

USD context (cost vs frontier): At $1.4/$4.4 per MTok, GLM-5.2 output is roughly one-sixth of Claude Opus 4.8 ($5/$25 per MTok) and well under half of Sonnet 5's introductory $2/$10. On Semgrep's IDOR benchmark, GLM-5.2 cost approximately $0.17 per real vulnerability found. Sources: Anthropic – Pricing , Semgrep – We Have Mythos At Home Glm 52 Beats Claude In Our Cyber Benchmarks

Batch API: 50% discount on supported GLM-4 series models (carried from May; not re-confirmed for GLM-5.2). Source: Bigmodel – Introduction

Search tools (carried from May): Search-Std ¥0.01/request, Search-Pro ¥0.03/request. Source: Bigmodel – Introduction

Model Performance / Benchmarks

GLM-5.2 published its own benchmark claims (partly as chart images), which third-party evaluators then independently tested.

Token efficiency caveat: GLM-5.2 uses 43K output tokens per Intelligence Index task (of which 37K is reasoning), up from GLM-5.1's 26K and above MiniMax-M3 (24K) and Kimi K2.6 (35K). Artificial Analysis places it off the most token-efficient quadrant for its intelligence level, so headline price-per-MTok advantages shrink somewhat in practice. Source: Artificialanalysis – Glm 5 2 Is The New Leading Open Weights Model On The Artificial Analysis Intelligence Index

Key context: Zhipu publishes some scores as exact numbers and others (BrowseComp, MCP-Atlas) only as "open-source SOTA" claims without figures, and several headline charts are images rather than extractable text. The GLM-5 generation uses a ~744B total / ~40B active MoE architecture and the open-source SLIME RL training framework. Source: Interconnects – Glm 52 Is The Step Change For Open

Latest News

GLM-5.2 Flagship Launch (June 13-16, 2026)

Zhipu rolled out GLM-5.2 to GLM Coding Plan members on Saturday, June 13, 2026, an unusually timed weekend release that landed days after the US export-control directive suspended Anthropic's Fable 5 / Mythos 5. The MIT-licensed open weights and official release blog followed on June 16. GLM-5.2 keeps the same ~744B/40B MoE size as GLM-5.1 but extends context to a "Solid 1M" lossless window, ships thinking + reasoning_effort controls, and is positioned for end-to-end project delivery (Zhipu cites a single task that consumed 850K+ tokens). Sources: Bigmodel – Glm 5.2 , Interconnects – Glm 52 Is The Step Change For Open

Off-Peak Multiplier Promotion Extended to September (June)

The Coding Plan page was updated so that GLM-5.2 and GLM-5-Turbo consume quota at 1x during off-peak hours through the end of September 2026, replacing the prior "through end of June" window. The permanent rate (3x peak 14:00-18:00 UTC+8, 2x off-peak) is unchanged. This confirms the May watch-item prediction ("reverts at end of June") did NOT come true. Source: Bigmodel – Overview

GLM-5.1 and GLM-5 Auto-Redirect to GLM-5.2 (June)

Calling the legacy GLM-5.1 or GLM-5 model identifiers now silently routes to GLM-5.2. All Coding Plan tiers ship GLM-5.2, GLM-5-Turbo, and GLM-4.7. Source: Bigmodel – Overview

Semgrep: GLM-5.2 Beats Claude Code on a Cyber Benchmark (June 22, 2026)

Security firm Semgrep ran open-weight and frontier models on an IDOR (insecure direct object reference) detection task with identical prompts. GLM-5.2 scored 39% F1, beating Claude Code (Opus 4.6, 37%; Opus 4.8/4.7, 28%) at roughly $0.17 per vulnerability found. It still trailed Semgrep's own multimodal pipeline (53-61% F1). Semgrep also flagged Zhipu's reward-hacking disclosure. Source: Semgrep – We Have Mythos At Home Glm 52 Beats Claude In Our Cyber Benchmarks

Artificial Analysis: GLM-5.2 is the Leading Open-Weight Model (June 16, 2026)

Artificial Analysis ranked GLM-5.2 first among open-weight models on its Intelligence Index v4.1 at 51, up 11 points over GLM-5.1, and on the Pareto frontier of intelligence vs cost per task. It placed level with GPT-5.5 (xhigh reasoning) on the agentic GDPval-AA v2 benchmark. Source: Artificialanalysis – Glm 5 2 Is The New Leading Open Weights Model On The Artificial Analysis Intelligence Index

Community Signals

HackerNews launch thread: GLM-5.2 arrives with genuine hype

The "GLM 5.2 Is Out" thread (772 points, 504 comments) and the Semgrep benchmark thread (1,099 points, 507 comments) made GLM-5.2 the most discussed open model of the month. The dominant sentiment was that GLM-5.2 has closed most of the gap to frontier closed models at a fraction of the cost. Sources: News – Item , News – Item

Developer experience quotes (with comment permalinks):

• wgd: "I've got a GLM subscription (mostly because I like supporting open model makers, pretty sure my monthly usage is so low that pay-per-token would be more cost effective)... I haven't noticed much of a capability gap between [GLM-5.1 and Opus] for the sorts of things I ask of an AI agent... If nothing else 1M context is a great improvement, feels like between DeepSeek v4, then MiniMax M3, and now GLM-5.2 adding it 1M is rapidly becoming 'table stakes' for agentic models." News – Item
• vcryan: "I've tried every version since 4.6 and this one is doing a great job a spec-implementation runner. If I had to guess... somewhere between Sonnet and Opus in terms of quality. Z.ai's issue has been service reliability. So far so good on day one." News – Item
• saratogacx: "I've been using GLM-5/5.1 for about 6 months and it has been a fairly capable model. I've seen a lot of mixed opinions that tend to align with harness usage so it is worth trying out a couple with a model before writing it off." News – Item
• vulture916: "It's gotten really good, just slow as all hell." News – Item

A counterpoint on workflow fit:

• maherbeg: "I've found the prompting needs are drastically different from the latest frontier models to the latest open weight models. I can be much more vague and talk about an end goal with the frontier models vs needing to be more prescriptive + have a workflow on the open weight models. This gap continues to close." News – Item

Cost-pressure and geopolitical framing

Several top comments framed GLM-5.2 alongside MiniMax M3 and Kimi K2.7 as a counter-narrative to the US export-control suspensions of Fable 5 / Mythos 5:

• evilturnip: "In the last few days, Chinese labs have given us MiniMaxM3, KimiK2.7 and now GLM5.2. Meanwhile US is censoring models. Reads like fiction." News – Item
• anonyfox: "If this goes open, distills are tried out, independent providers around the world host it with actual price competition, the house of cards for anthropic collapses pre-ipo. The floor is opus (open models caught up), the current ceiling is Mythos (self inflicted ban due to the safety bullshit theater)." News – Item

Analyst and vendor reception

• Nathan Lambert (Interconnects): "GLM-5.2 is the open weight model that feels right in coding harnesses as a general agent. It's the first one." He compared the release to the DeepSeek R1 moment and dated the open-closed gap at ~204 days (~6.8 months) behind Opus 4.5. Interconnects – Glm 52 Is The Step Change For Open
• Vercel CEO Guillermo Rauch: "Genuinely impressed, almost shocked, at how good GLM-5.2 by @zai_org is at coding. This changes things." X – Status

English-language community is no longer sparse

Unlike the May report, which found minimal English-language discussion, GLM-5.2 generated large HackerNews threads, multiple independent benchmark posts (Semgrep, Artificial Analysis, ThinkWright, techstackups), and an Unsloth local-run guide (615 points). The English-language community footprint for Zhipu is now substantial. Sources: News – Item , Artificialanalysis – Glm 5 2

Enterprise Readiness

Terms explained:

• SSO (SAML) - employees log in via a corporate identity provider (Okta, Azure AD) instead of separate passwords. Not documented for Zhipu.
• SCIM - automated user provisioning/deprovisioning from a corporate directory. Not documented for Zhipu.
• IP indemnity - the provider covers legal costs if its AI output infringes a third party's copyright. Not documented for Zhipu.

Transparency Gaps