Tier A — Frontier
Runs in:CNMade in:China
Z.ai (GLM / Zhipu)

GLM-5

Tier A — Frontier · 205K tokens

Tokonomix Editorial Team·Reviewed by Mes Kalkan··

GLM-5 is the base model of Zhipu’s GLM-5 generation on z.ai — the generation’s workhorse, priced below the GLM-5.2 flagship. It is an OpenAI-compatible reasoning chat model with tool and JSON support.

Section 01

Speed analysis

Latency measured across all benchmark runs. P50 (median) and P95 (95th percentile) give a realistic picture of response speed under normal and peak load.

P50 latency (median)P95 latency105 runs
1702853615371222052903909-2610-22ms
Section 02

Quality scores

How this model compares to the rest of the field on each prompt category, from a pairwise fit over the same prompts. The raw judge score sits underneath each number.

58%
Coding
judge mean 94
88%
Creative
judge mean 98
63%
Factual
judge mean 93
63%
Reasoning
judge mean 97

Win rate per category: how often this model beats a field-average model on a prompt from that category. 50% is average, not a failing grade. It is not a percentage of correct answers.

Section 03

Pricing

What you pay per million tokens when you use this model on Tokonomix, plus an estimate for a typical conversation.

💰
API rates — GLM-5
$1.50 per 1M input tokens
$4.79 per 1M output tokens
≈ $0.0019 per typical conversation (800 tokens)
Input vs output price (per 1M tokens)
per 1M input tokens$1.50
per 1M output tokens$4.79
Section 04

Tokens per second

Throughput in tokens per second, derived from measured P50 latency. Higher is better; fluctuations track provider-side load.

Throughput (tokens / s)66 / avg 72
1178

Estimated from P50 latency × 200 output tokens — the absolute number depends on this assumption; the trend is what matters.

Section 05

Capabilities

jsonnotes: GLM emits a non-standard reasoning_content field beside content; read content for the answer.toolsreasoning
Section 06

Availability

Availability

How often this model answers when we call it — measured across real API requests and live tests over the last 30 days. This is separate from quality: these numbers only tell you whether the model responds, not how good the answer is.

Last 7 days

—

Last 30 days

50.0%

n=2

Median response time

2,465ms

n=1

Based on 377 measurements over the last 30 days.

Technical details

Only live API calls and live-test requests count — internal probes and benchmark runs are excluded.

Calls with a custom API key (BYOK) are excluded: those failures are key-specific, not a sign of model downtime.

Failed calls are NOT included in quality scores — quality is measured on successful responses only. Availability and quality are independent signals.

Median response time (p50) across successful calls with a recorded duration. Outliers (very slow or very fast calls) pull the median less than the average.

Total calls (30d)

2

OK responses (30d)

1

Total calls (7d)

0

OK responses (7d)

0

Section 07

Tokonomix benchmark verdicts

⚖️
Endorsed by 1 judge
Independent LLM judges evaluated this model on our weekly intelligence tests
claude-sonnet-4-595/100 · 34 runs
32 correct0 partial2 wrong94% accuracy
● 2026-10-18

GLM-5 shows quality regression alongside notable latency improvements

GLM-5 demonstrates a mixed performance shift in this benchmark window, with a concerning 6-point drop in overall quality score from 100 to 94, while simultaneously achieving a significant 36% improvement in response latency. The factual accuracy category, which represents the sole evaluated dimension in both windows, declined from a perfect 100 to 94, indicating potential regression in core reasoning or information retrieval capabilities. The latency improvement from 7575ms to 4878ms represents meaningful progress in response speed, bringing median latency below 5 seconds for the first time in observed windows. This enhancement suggests underlying infrastructure optimizations or model efficiency improvements that benefit user experience through faster interactions. However, the quality degradation warrants attention from users who prioritize accuracy over speed. The shift from perfect factual performance to a still-strong but imperfect 94 suggests possible trade-offs made during optimization, or changes in model behavior that impact correctness. With only one test run per window, these results should be interpreted as initial signals rather than definitive trends. Users should monitor whether subsequent windows confirm sustained latency gains while quality stabilizes or recovers to previous levels.

Quality

94.0

Latency p50

4,878 ms

Test runs

1

✗ Quality dropped 6 points✗ Factual accuracy declined to 94✓ Latency improved 36%✓ Response speed under 5 seconds
Section 08

Full model profile

GLM-5: the base of Zhipu’s newest generation

GLM-5 is the base model of Zhipu’s GLM-5 generation on z.ai — the generation’s workhorse, priced below the GLM-5.2 flagship. It is an OpenAI-compatible reasoning chat model with tool and JSON support.

We registered it with a large (~200K-token) context window as a provisional figure; GLM-5-generation documentation is thin, so confirm the exact window on z.ai before relying on it.

Architecture & training signals

GLM-5 is the base of Zhipu AI’s GLM-5 generation. Public technical detail is still limited as of July 2026; we treat it as a large reasoning-oriented chat model with tool-calling and structured JSON over an OpenAI-compatible endpoint. Like the rest of the GLM line, it returns a non-standard reasoning_content field alongside content in its OpenAI-compatible responses; integrations should read content for the final answer and treat reasoning_content as an optional trace.

Where it shines

  • A balanced quality-vs-price point within the newest GLM generation.
  • Long-context reasoning and analysis.
  • Cross-family diversity in a consensus panel.

Where it falls short

  • Still pricier than the GLM-4.x line and the free flash tiers.
  • Limited reproducible benchmark data; verify on your workload.
  • Non-EU hosting.

Real-world use cases

  • General reasoning where you want current-generation GLM quality without the flagship price.
  • Document analysis and summarisation.
  • Agentic tool-use pipelines.

Tokonomix benchmark snapshot

GLM-5 is newly registered on Tokonomix and not yet activated, so we have not run it through our weekly intelligence test or speed benchmark. There are no Tokonomix scores to report yet — and we will not invent any.

When it goes live, it enters the same weekly harness as every other model: identical prompts, an independent cross-family judge, and reproducible latency and cost measurements. Until then, treat the pricing and capability notes on this page as the vendor-published starting point, not as measured Tokonomix results.

EU privacy & data residency

GLM-5 is built by Zhipu AI (z.ai), a China-headquartered lab, and is served from non-EU infrastructure. This is important to state plainly: routing a prompt to this model is not an EU-data-residency or GDPR-sovereign choice, and Tokonomix will never tag it as one.

If your use case requires data to stay within the EU, pick a model whose provider is EU-hosted (for example our OVH or Azure-EU routes) rather than a GLM model. Tokonomix keeps z.ai out of every EU-only / sovereign routing set by design. Use GLM where its capability or price is the priority and cross-border processing is acceptable for that workload.

Verdict & alternatives

GLM-5 is the sensible default within the GLM-5 generation: most of the newest-generation quality at a lower price than GLM-5.2. If you need the absolute top, go GLM-5.2; if budget dominates, drop to GLM-4.6/4.5 or the free flash models.

Last automated test
Oct 22, 2026 · 08:01 UTC · Speed benchmark
P50 latency
3053 ms
P95 latency
3511 ms
Errors
0 / 6 runs
Last reviewed by Tokonomix Team·September 15, 2026