# zai-glm-53-emergent-cyber-2026-08-14

## Veille

Announcement post published on the **official Z.ai blog** (formerly Zhipu AI, Chinese lab) on **August 14, 2026**, **with no individual byline**, ~2,000 words plus footnotes. It announces **GLM-5.3**, successor to GLM-5.2, opening with a methodological thesis: *« Scaling post-training is all we did for GLM-5.3. »* Same base model as GLM-5.2 — *« every gain comes from post-training »*. Three announcements. **(A) An open-weights coding model**: +50% claimed on **Z.ai Code Bench**, an unpublished in-house benchmark. **(B) A cyber capability presented as "emergent"**, which the body of the text traces to a training choice — *« As part of post-training, we introduced vulnerability discovery data and environments into the training mix. We expected this to make the model better at finding and reasoning about vulnerabilities »* — what came as a surprise was the speed and the change in nature: the model moves from identifying isolated flaws to *« coherent plans for complete exploitation chains »*. Gains grow with position in the exploitation chain: CyberGym 77.2 → **84.5%**, ExploitBench 24.4 → **54.4%** (×2.2), ExploitGym 29 → **105** tasks in 2h (×3.6), with the gap to the closed frontier remaining wide (181 and 247 tasks). Z.ai puts it this way: *« Capability is growing fastest exactly where we are furthest behind. »* The post also publishes a **Z.ai Security Disclosure Ledger**: **2,436 vulnerabilities identified across 269 open source projects** — kernels, OSes, browser engines, infrastructure, web applications, network protocols — the oldest introduced in **1981**, average lifetime before discovery **26.6 years**, of which **53 disclosed** and **2,383 under embargo**. **(C) A weight release** *« within two weeks of launch, once safety evaluation and hardening are complete »*. The most reusable methodological contribution: **environment and verifier synthesis**, the latter produced without access to the reference solution and admitted only after a triptych of negative controls — **oracle**, **no-op**, **unsolved-state**. All agentic evaluations are conducted **in Claude Code 2.1.207**.

## Titre Article

GLM-5.3: Frontier Coding with Emergent Cyber Capabilities

## Date

2026-08-14

## URL

https://z.ai/blog/glm-5.3

## Keywords

GLM-5.3, GLM-5.2, Z.ai, Zhipu AI, open weights, open-weights, Chinese lab, post-training, post-training scaling, RL scaling, environment synthesis, long-horizon tasks, SAO, IndexShare, slime, Megatron, SGLang, multi-teacher OPD, synthetic verifier, verifier synthesis, binary reward, reward hacking, oracle control, no-op control, unsolved-state control, judge agent, human-in-the-loop, cyber capability, vulnerability discovery, exploitation chain, CyberGym, ExploitBench, ExploitGym, coordinated disclosure, Security Disclosure Ledger, embargo, 2436 vulnerabilities, 269 projects, 2383 under embargo, 1981, 26, 6 years, dual-use, weight release, safety evaluation, hardening, Terminal-Bench 3.0, DeepSWE, SWE-Marathon, PostTrainBench, Agents' Last Exam, GDPval-AA v2, Z.ai Code Bench, private benchmark, benchmark contamination, token efficiency, reasoning_effort, Claude Code 2.1.207, evaluation harness, Artificial Analysis, Kimi K3, GLM Coding Plan, off-peak pricing, ZCode

## Authors

**Z.ai** (anciennement **Zhipu AI**), laboratoire d'IA chinois, éditeur de la famille **GLM**. Billet **institutionnel et non signé** : aucun auteur nommé, aucun chercheur mis en avant, aucun lien vers un rapport technique ou une carte de modèle. Publié le **14 août 2026**. La page est une SPA React — le HTML servi est un `<div id="root">` vide, et le texte comme les scores ont dû être extraits du bundle `glm-5.3-BCnx8T5_.js`, où ils figurent en valeurs source.

## Ton

**Profile**: product announcement from a model lab, register of an **abbreviated technical report** — dense, quantified, free of marketing superlatives. Audience: ML engineers, platform teams, API buyers. An engineers' note whose only marketing element is the title.

**Style**: the opening is a methodological thesis, not a performance claim — *« Scaling post-training is all we did for GLM-5.3 »*, first sentence, bolded: the announcement says less "here is a better model" than "here is what we found most cost-effective to do." The admission of lagging behind is left in plain view, unpadded: *« Mythos 5 remains well ahead at 181 and 247 tasks »*, then *« Capability is growing fastest exactly where we are furthest behind. »* The title uses *« emergent »* where the body describes a deliberate addition of vulnerability discovery data to the training mix; the only element the text calls unexpected is the pace. The method description is more generous than the sector average — the environment and verifier synthesis chain is laid out step by step and closes on an acknowledged limitation: *« These pipelines still require a meaningful amount of human-in-the-loop work. »* The footnotes carry the operational truth: temperatures, context windows, timeouts, avg@3, anti-cheat whitelists, and above all the **modifications made to the official harnesses** (anti-cheat checks removed on `strip-clone` for false positives, `--extra-index-url` added on `parameter-golf` and `trimul-cuda` to fix Docker builds) — a rare disclosure, which in turn makes the scores incomparable to official leaderboards.

**Marker phrases**:
- ***« Scaling post-training is all we did for GLM-5.3 »***
- ***« every gain comes from post-training »***
- ***« much of the difficulty in scaling post-training moves from the model to the environment »***
- ***« a judge agent then attempts each task to verify that it is actually solvable »***
- ***« Verifiers are synthesized without access to the reference solution »***
- ***« it began to reason across multiple stages of exploitation, forming coherent plans for complete exploitation chains »***
- ***« Capability is growing fastest exactly where we are furthest behind »***
- ***« We will release the weights in two weeks after launch, once safety evaluation and hardening are complete »***
- ***« Disabling thinking is no longer supported »***

**Epistemic position**: an interested party, transparent about its method, judging its own performance. Reliable on *how* the model was trained and *under what conditions* it was measured; unverifiable on *how much* it is worth — the central claim rests on a private benchmark and competitor scores were produced by Z.ai, with three attributed exceptions (GDPval-AA v2 and FrontierSWE to **Artificial Analysis**, Toolathlon Verified via the official evaluation service).

## Pense-betes

- **Date / source**: **August 14, 2026**, official **Z.ai** blog, unsigned post. Page built as a React SPA, not capturable by an ordinary `curl | lynx`; bundle name is hashed, hence unstable.
- **Key framing**: same base model as GLM-5.2, all gains coming from post-training. The term *« emergent »* in the title refers, in the body, to a **rate of progress** — the capability itself having been deliberately trained. ### The cyber curve: the gain grows with position in the chain | Benchmark | What it measures | GLM-5.2 → GLM-5.3 | Factor | Best in table | |---|---|---|---|---| | CyberGym | identifying and validating a flaw in white-box | 77.2 → **84.5%** | ×1.09 | **GLM-5.3** (ahead of 83.8 / 83.6) | | ExploitBench | reasoning about exploiting real flaws | 24.4 → **54.4%** | **×2.2** | 78.0 (closed model) | | ExploitGym (2h) | exploitation tasks completed, normalized budget | 29 → **105** | **×3.6** | 181 (closed model) | | ExploitGym (6h) | same, 6h | 39 → **130** | ×3.3 | 247 (closed model) | Two simultaneous readings, both accurate: Z.ai remains far behind the closed frontier everywhere exploitation becomes real, and that is exactly where its progress is fastest — for one additional month of post-training on the same base model. Citation trap: ExploitGym budgets are **throughput-normalized**, API time being rescaled by each model's TPS (115 for GLM-5.3, 40 for Kimi K3, 47 for Qwen3.8-Max, sourced from Artificial Analysis). These are not wall-clock hours. ### The Security Disclosure Ledger | Counter | Value | Share | |---|---|---| | Findings tracked | **2,436** | 100% | | Publicly disclosed | **53** | **2.2%** | | Under embargo | **2,383** | **97.8%** | | Critical & High | 1,097 | 45% | | OSS projects affected | **269** | — | An embargo falls under ordinary coordinated disclosure; the size of the backlog and the publication rate are what stand out. Four points the post does not address: the delay policy, actual notification of the 269 projects' maintainers, the identity of the *« several security teams in China »* mentioned, and what happens to the embargo once the weights are released — a public model able to rediscover these flaws effectively renders the embargo moot. ### The weight release timeline *« We will release the weights in two weeks after launch, once safety evaluation and hardening are complete. »* What is not specified: the definition of *hardening* for open weights, the criteria that would trigger a delay, the existence of a named third-party evaluator or red team, and any commitment to publish the report. The timeline is anchored to the commercial launch, not to the completion of a process. Practical consequence: as of August 14, 2026, GLM-5.3 is an API model paired with an intention to open up, not to be built into a reversibility architecture before actual release. Compare with the treatment of weight release in [[zuckerberg-meta-future-is-for-everyone-superintelligence-2026-08-10]]. ### The claim "most capable open-weights model for coding" Across the 7 benchmarks in the *Coding* group where **Kimi K3** is reported in Z.ai's table: | Benchmark | GLM-5.3 | Kimi K3 | Edge | |---|---|---|---| | Terminal Bench 2.1 | 88.2 | **88.3** | Kimi K3 | | Terminal-Bench 3.0 | **28.3** | 17.4 | GLM-5.3 (large) | | DeepSWE (v1.1) | 66.9 | **67.5** | Kimi K3 | | NL2Repo | 58 | 58 | tie | | ProgramBench | **19** | 17.5 | GLM-5.3 | | SWE-Marathon (v1.1) | 42.5 | **48.1** | Kimi K3 (large) | | PostTrainBench | **39.8** | 32 | GLM-5.3 | That is **3–3 with one tie**. Two narrower claims do hold up, however: *« open-source SOTA on Terminal Bench 3.0 »* (28.3 vs 17.4) and on *Agents' Last Exam* (28.5 vs 27.6 for Kimi K3 and 27 for Qwen3.8-Max, within 0.1 point of the best closed model in the table). ### The gain from one generation of post-training | Benchmark | GLM-5.2 | GLM-5.3 | Gap | |---|---|---|---| | Terminal-Bench 3.0 | 4.6 | **28.3** | **×6.2** | | SWE-Marathon (v1.1) | 19.4 | **42.5** | ×2.2 | | DeepSWE (v1.1) | 46.2 | **66.9** | +20.7 pt | | AutomationBench | 26.2 | **48.2** | +22 pt | | Agents' Last Exam | 23.8 | **28.5** | +4.7 pt | | GDPval-AA v2 | 1508 | **1769** | +261 | Reading: on long agentic tasks, the training environment turns out to be a more cost-effective lever than fresh pretraining — *« much of the difficulty in scaling post-training moves from the model to the environment »*. A ×6.2 gap in a single month also indicates the benchmark was far from saturated: Terminal-Bench 3.0 still sits at 28.3%, i.e. more than seven tasks in ten failing. Context on agentic benchmarks: [[patel-block-buzz-teams-tokens-benchmarks-2026-08-06]]. ### Cost per successful task (Z.ai Code Bench, private) | Configuration | Score | Output tokens / task | |---|---|---| | GLM-5.2, Max effort | 23.4% | 96K | | **GLM-5.3, Max effort** | **34.5%** | **~75K** | | GLM-5.3, High effort | 31.4% | ~50K | | Claude Opus 4.8 (Z.ai measurement) | 29.5% | 120K | | Claude Fable 5, Max effort | **39.5%** | not disclosed | The selling point is the ratio: +5 points over Opus 4.8 for roughly 1.6× fewer output tokens. Three caveats to keep together: in-house and private benchmark, self-adjudicated comparison, and Fable 5 still stays ahead — Z.ai publishes it. The cost of output tokens is also not the total cost (input, cache, number of turns). ### The verification chain, transposable outside Z.ai 1. Research agents collect task patterns from real work and convert them into executable, long-horizon environments with multi-step dependencies and hidden state. 2. A **judge agent attempts each task** to verify it is actually solvable. 3. **Verifiers are synthesized without access to the reference solution** — so they cannot encode the answer. 4. Solver trajectories are used to discover and close reward shortcuts. 5. A verifier is admitted only if it passes **oracle** (accepts the true solution), **no-op** (rejects doing nothing) and **unsolved-state** (rejects the initial state). Only then does it *« produce a binary reward reliable enough to train on directly »*. Points 3 and 5 transpose to any automatic test generation: a test that doesn't fail red on the original code proves nothing. Same requirement as formulated in [[dumortier-marketing-ai-os-verification-2026-08-12]]. Limitation acknowledged by Z.ai: *« These pipelines still require a meaningful amount of human-in-the-loop work. »* ### Claude Code as evaluation harness Seven evaluations documented in footnotes are conducted *« in Claude Code 2.1.207 »* — Terminal-Bench 2.1 and 3.0, ALE, CyberGym, ExploitGym, ExploitBench, PostTrainBench, SWE-Marathon. A Chinese lab measures its frontier model and those of its competitors there, citing the version as a dependency. Uncounted corollary: the transferability of scores to another harness. ### API break and pricing
- *« Disabling thinking is no longer supported by GLM-5.3 »*: `thinking.type: "disabled"` causes the request to fail. Prescribed migration: switch to `enabled` **and** `reasoning_effort: "low"` **before** changing the model ID. Three levels `low` / `high` / `max`, **default `max`**. The floor cost of a call therefore rises; for any high-volume, low-complexity workload, the switch to `low` must precede the ID change.
- **GLM Coding Plan** uses point quotas, counted separately for input, cached input, and output. **Peak hours 14:00–18:00 UTC+8, Monday to Friday**, everything else at 50% of the rate — meaning, for a European team, a peak window corresponding to 08:00–12:00 Paris time. Associated ZCode offering: claimed *« 98%+ cache hit rate »*, ~30% more effective tokens, 1.5× quota boost through **August 31, 2026**, *Goal* mode and *Remote Control* via **WeChat or Feishu**. ### Three internal inconsistencies, not to be propagated 1. *« 1,097 medium-to-high severity issues »* contradicts the same post's own sidebar: critical 107 + high 990 = **1,097**, medium 1,286 separate. Cite **"1,097 critical or high-severity flaws"**. 2. *« the oldest dating back roughly 40 years »* contradicts the sidebar (*« introduced in 1981 »*, *« 45 YEARS OF IMPACT »*). Use **1981** and an average lifetime of **26.6 years**. 3. The body attributes to **"Mythos 5"** the scores of 83.8 / 78.0 / 181-247 that the table attributes to **"Fable 5 (w/ fallback)"**. These are two distinct Anthropic models and the post does not resolve which. Do not attribute these scores to a named model. ### Training infrastructure Post-training on **slime** (Z.ai's open source framework), **Megatron** for training, **SGLang** for rollout, with training, rollout, and buffer on a single data stream — environments plug in like data generation, not as loop modifications. Claimed gains: average training/rollout log-probability gap controlled to **1e-7**, **multi-teacher OPD** with dynamic switching and prefetching, joint router/slime scheduling, for a total of **more than 2.3× end-to-end throughput** on long-horizon coding RL tasks. Underlying publications: **IndexShare** (arXiv 2603.12201) and **SAO** (arXiv 2607.07508). ### Citation hygiene
- Everything from **Z.ai Code Bench** is unverifiable by construction: cite it as *« per an unpublished internal benchmark »*.
- Every competitor score in the table was produced by Z.ai, except GDPval-AA v2 and FrontierSWE (Artificial Analysis) and Toolathlon Verified (official service).
- The public harnesses were modified; the scores are not comparable to official leaderboards.
- The link to the sovereignty file is direct: three days earlier, [[nunez-mistral-gigawatt-compute-europeen-venturebeat-2026-08-11]] announced that Mistral AI hosts GLM-5.2 under European regional controls. The safety evaluations Mistral says it runs on third-party models have not been publicly described.

## RésuméDe400mots

Announcement post published on **August 14, 2026** on the **Z.ai** blog (formerly Zhipu AI), **unsigned**, for the launch of **GLM-5.3**.

**The methodological thesis.** *« Scaling post-training is all we did for GLM-5.3. »* Same base model as GLM-5.2: **all the gain comes from post-training**, built on the stack from the previous cycle — **IndexShare** (long context), **SAO** (long-horizon RL) and **slime** (asynchronous training, Megatron + SGLang). The bottleneck has shifted from the model to **the environment**: Z.ai describes pipelines that **synthesize** environments and reward signal — a judge agent verifies solvability, **verifiers are synthesized without access to the reference solution**, and are admitted only after a triptych of **oracle / no-op / unsolved-state** controls. The work remains *« human-in-the-loop »*. End-to-end RL throughput improved by **more than 2.3×**.

**The coding results.** Terminal-Bench 3.0 goes from **4.6 to 28.3**, DeepSWE v1.1 from **46.2 to 66.9**, Agents' Last Exam from **23.8 to 28.5**. On **Z.ai Code Bench**, an **in-house, private** benchmark, +50% over GLM-5.2, with a simultaneous gain in **token efficiency**: 34.5% at ~75K output tokens at Max effort (versus 23.4% at 96K for GLM-5.2), and 31.4% at ~50K at High effort — ahead of Claude Opus 4.8 (29.5% at 120K). **Claude Fable 5 remains ahead at 39.5%.** The claim *« most capable open-weights model for coding »* **does not follow from the table**: against **Kimi K3**, the score is **3–3 with one tie**.

**The cyber capability.** Presented as *« emergent »*, it was **deliberately trained** — the post writes *« we expected this to make the model better »*. What came as a surprise was the **speed**, and the shift from isolated flaws to the **complete exploitation chain**. CyberGym **84.5%** (best in the table), ExploitBench **54.4%** (×2.2), ExploitGym **105/130 tasks** (×3.6 over GLM-5.2, throughput-normalized budgets). Key sentence: ***« Capability is growing fastest exactly where we are furthest behind. »***

**The heaviest number.** Working with Chinese security teams, the model identified **2,436 vulnerabilities in 269 open source projects** — kernels, OSes, browser engines, network protocols — the oldest introduced in **1981**, average lifetime **26.6 years**. The **Security Disclosure Ledger** shows **53 disclosed** and **2,383 under embargo**: **2.2% published**.

**Governance.** Weights announced *« in two weeks, once safety evaluation and hardening are complete »* — **a date, not a criterion**: no definition of hardening, no condition for non-release, no third-party evaluator.

**Miscellaneous.** `thinking.type: "disabled"` **is no longer supported** (migration required); GLM Coding Plan quotas in points, **50% outside 14:00–18:00 UTC+8**; **nearly all evaluations are conducted in Claude Code 2.1.207**.

## GrapheDeConnaissance

- Z.ai —publie→ GLM-5.3 (TECHNOLOGIE, 0.98)
- GLM-5.3 —est_variante_de→ GLM (TECHNOLOGIE, 0.97)
- Z.ai —publie→ GLM (TECHNOLOGIE, 0.98)
- GLM-5.3 —remplace→ GLM-5.2 (TECHNOLOGIE, 0.97)
- GLM-5.3 —est_basé_sur→ GLM-5.2 (TECHNOLOGIE, 0.96)
- Z.ai —affirme_que→ GLM-5.3 utilise le même modèle de base que GLM-5.2 et que la totalité du gain provient du post-entraînement — « Scaling post-training is all we did » (CITATION, 0.97)
- GLM-5.3 —utilise→ slime (TECHNOLOGIE, 0.95)
- slime —permet→ de brancher maths, code, sandboxes, vérificateurs et environnements agentiques long-horizon comme de la génération de données plutôt que comme des modifications de la boucle d'entraînement, entraînement, rollout et buffer tenant sur un seul flux de données (AFFIRMATION, 0.93)
- slime —améliore→ le débit d'entraînement RL end-to-end de plus de 2,3× sur les tâches de codage long-horizon, avec un écart moyen de log-probabilités entraînement-rollout contrôlé à 1e-7 (MESURE, 0.9)
- GLM-5.3 —utilise→ SAO (METHODOLOGIE, 0.92)
- GLM-5.3 —utilise→ IndexShare (TECHNOLOGIE, 0.9)
- Z.ai —affirme_que→ la difficulté du passage à l'échelle du post-entraînement se déplace du modèle vers l'environnement, un environnement utile devant être exécutable, vérifiable et proche du travail professionnel réel, et devant exister en grand nombre (AFFIRMATION, 0.95)
- synthèse d'environnements de tâches —permet→ de produire des environnements long-horizon exécutables et leur signal de récompense à partir de motifs collectés sur du travail réel, un agent juge vérifiant au préalable que chaque tâche est effectivement résoluble (AFFIRMATION, 0.93)
- synthèse d'environnements de tâches —recommande→ de synthétiser les vérificateurs sans accès à la solution de référence et de ne les admettre qu'après trois contrôles négatifs — oracle, no-op et unsolved-state — afin d'obtenir une récompense binaire assez fiable pour entraîner directement dessus (AFFIRMATION, 0.94)
- synthèse d'environnements de tâches —réduit→ les raccourcis de récompense, les trajectoires de solveur servant à les découvrir puis à les fermer (AFFIRMATION, 0.9)
- GLM-5.3 —mesure→ 28,3 sur Terminal-Bench 3.0 contre 4,6 pour GLM-5.2, 66,9 sur DeepSWE v1.1 contre 46,2, et 28,5 sur Agents' Last Exam contre 23,8 (MESURE, 0.96)
- Z.ai —a_créé→ Z.ai Code Bench (TECHNOLOGIE, 0.95)
- Z.ai Code Bench —réduit→ le risque de contamination par les jeux de test publics, au prix de l'impossibilité pour un tiers de répliquer la mesure (AFFIRMATION, 0.88)
- GLM-5.3 —mesure→ 34,5 % sur Z.ai Code Bench à environ 75 000 tokens de sortie par tâche en effort Max, contre 23,4 % à 96 000 pour GLM-5.2, et 31,4 % à environ 50 000 tokens en effort High (MESURE, 0.93)
- GLM-5.3 —surpasse→ Claude Opus 4.8 (TECHNOLOGIE, 0.85)
- Z.ai —affirme_que→ GLM-5.3 est le modèle à poids ouverts le plus capable pour le codage, revendication que son propre tableau ne départage pas face à Kimi K3, à égalité sur les benchmarks de codage renseignés pour les deux (AFFIRMATION, 0.86)
- GLM-5.3 —concurrence→ Kimi K3 (TECHNOLOGIE, 0.92)
- Z.ai —affirme_que→ la capacité cyber s'est développée plus vite qu'attendu alors que les données et environnements de découverte de vulnérabilités avaient été délibérément introduits dans le mélange d'entraînement, le résultat visé étant précisément d'améliorer le modèle sur ce terrain (AFFIRMATION, 0.95)
- GLM-5.3 —permet→ de raisonner à travers plusieurs étapes d'exploitation et de former des plans cohérents pour des chaînes d'exploitation complètes, au lieu d'identifier des failles isolées (AFFIRMATION, 0.93)
- GLM-5.3 —mesure→ 84,5 % sur CyberGym contre 77,2 % pour GLM-5.2, 54,4 % sur ExploitBench contre 24,4 %, et 105 puis 130 tâches ExploitGym sous budgets normalisés de 2 h et 6 h contre 29 et 39 (MESURE, 0.95)
- Z.ai —affirme_que→ plus un benchmark se situe haut dans la chaîne d'exploitation, plus le gain sur GLM-5.2 est grand et plus l'écart au frontier fermé reste large — « Capability is growing fastest exactly where we are furthest behind » (CITATION, 0.96)
- GLM-5.3 —observé_dans→ 2 436 vulnérabilités identifiées dans 269 projets open source après revue d'experts, tri et déduplication, couvrant noyaux système, systèmes d'exploitation, moteurs de navigateur, infrastructure open source, applications web et protocoles réseau (MESURE, 0.93)
- Z.ai —a_créé→ Z.ai Security Disclosure Ledger (TECHNOLOGIE, 0.94)
- Z.ai Security Disclosure Ledger —mesure→ 2 436 constats suivis dont 53 publiquement divulgués et 2 383 sous embargo, 1 097 de sévérité critique ou élevée, la plus ancienne faille introduite en 1981 et une durée de vie moyenne de 26,6 ans avant découverte (MESURE, 0.94)
- Z.ai —collabore_avec→ plusieurs équipes de sécurité en Chine, non nommées, pour exécuter les modèles GLM contre des bases de code réelles depuis GLM-5.2 (AFFIRMATION, 0.9)
- Z.ai —prédit→ la publication des poids de GLM-5.3 dans les deux semaines suivant le lancement, une fois l'évaluation de sûreté et le durcissement terminés (AFFIRMATION, 0.95)
- publication de poids ouverts —s_applique_à→ un calendrier annoncé plutôt qu'à une procédure publiée : ni définition du durcissement, ni critère de non-publication, ni évaluateur tiers ne sont fournis (AFFIRMATION, 0.88)
- GLM-5.3 —observé_dans→ Claude Code (TECHNOLOGIE, 0.94)
- Claude Code —est_instance_de→ harnais d'évaluation agentique de référence, utilisé en version 2.1.207 par un laboratoire concurrent pour mesurer son propre modèle et ceux d'Anthropic et d'OpenAI (AFFIRMATION, 0.89)
- Artificial Analysis —mesure→ les scores GDPval-AA v2 et FrontierSWE du tableau comparatif, seules évaluations attribuées à un tiers, ainsi que les débits par modèle servant à normaliser les budgets ExploitGym (MESURE, 0.92)
- GLM-5.3 —s_oppose_à→ la désactivation du raisonnement : thinking.type « disabled » n'est plus supporté et fait échouer la requête, trois niveaux low, high et max étant proposés avec max par défaut (AFFIRMATION, 0.95)
- GLM Coding Plan —utilise→ un système de quotas en points comptés séparément pour l'entrée, l'entrée en cache et la sortie, les appels hors heures pleines — hors 14h-18h UTC+8 du lundi au vendredi — consommant 50 % du barème (AFFIRMATION, 0.93)
- ZCode —permet→ un taux de cache revendiqué à plus de 98 %, un mode Goal qui planifie, code, teste et vérifie jusqu'à atteinte de la cible, et un pilotage à distance des tâches longues via WeChat ou Feishu (AFFIRMATION, 0.9)

---
Canonical: https://www.thekb.eu/en/fiches/zai-glm-53-emergent-cyber-2026-08-14/