Signal SentrySearchUpdated Nov 20, 05:30 UTCAM Drop
Models · scorecard

GLM 5.3

What the vendor claims, next to what independent boards measured. The two are never mixed in one column.

At a glance

Maker
Z.ai
Price per 1M tokens
$0.0699 in · $1.4 out
Context window
1.05M
Input → output
text → text
Added to OpenRouter
Aug 18, 2026

Source: OpenRouter, as of Nov 20, 06:00 UTC. Model page on OpenRouter

What the vendor claims

Quoted word for word from the vendor's own launch post or model card. Claims, not measurements.

No vendor claims recorded yet. They come from launch posts and model cards the Drops covered.

Independent results

Measured by independent sources (Epoch AI, LMArena, OpenRouter's benchmark runs, τ²-bench, Vectara), not by the vendor. "Listed as" is the source's own name for the model, so you can check our match.

Independent benchmark results for this model, each with its source and date
BoardResultRank
LMArena · Text (style control)Listed as: GLM-5.3 (Max)LMArena (CC BY 4.0), as of Nov 18, 2026. Source1,483Arena score29
OpenRouter · GPQA DiamondListed as: GLM 5.3via Z.AI, its own providerSource: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Nov 20, 2026. Source87.0%Accuracy60
OpenRouter · GPQA Diamond: cost per taskListed as: GLM 5.3via Z.AI, its own providerLower is better on this board.Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Nov 20, 2026. Source$0.10Cost per task36
OpenRouter · GPQA Diamond: time per taskListed as: GLM 5.3via Z.AI, its own providerLower is better on this board.Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Nov 20, 2026. Source352 sTime per task49
OpenRouter · τ²-bench airlineListed as: GLM 5.3via Z.AI, its own providerSource: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Nov 20, 2026. Source77.8%Accuracy13
OpenRouter · τ²-bench airline: cost per taskListed as: GLM 5.3via Z.AI, its own providerLower is better on this board.Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Nov 20, 2026. Source$0.082Cost per task24
OpenRouter · τ²-bench airline: time per taskListed as: GLM 5.3via Z.AI, its own providerLower is better on this board.Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Nov 20, 2026. Source136 sTime per task19

Price history

OpenRouter's list price per 1M tokens in each of our snapshots.

  • Output now $1.4
  • Input now $0.0699

Changed 24 times across our 93 snapshots.

Show the numbers
This model's OpenRouter price in each of our snapshots
SnapshotInput $/1MOutput $/1M
Nov 20, 06:04 UTC$0.0699$1.4
Nov 19, 16:04 UTC$0.0699$1.4
Nov 19, 04:04 UTC$0.0699$1.4
Nov 18, 18:04 UTC$0.0699$1.4
Nov 18, 04:03 UTC$0.0699$1.4
Nov 17, 16:01 UTC$0.0699$1.4
Nov 17, 04:04 UTC$0.0699$1.4
Nov 16, 19:54 UTC$0.0699$1.4
Nov 16, 04:05 UTC$0.0699$1.4
Nov 15, 16:04 UTC$0.084$3.6
Nov 15, 04:03 UTC$0.084$3.6
Nov 14, 16:04 UTC$0.084$3.6
Nov 14, 04:05 UTC$0.084$3.6
Nov 13, 16:05 UTC$0.084$3.6
Nov 13, 04:04 UTC$0.084$3.6
Nov 12, 16:05 UTC$0.0699$1.4
Nov 12, 04:04 UTC$0.0699$1.4
Nov 11, 16:01 UTC$0.0699$1.4
Nov 11, 04:04 UTC$0.0699$1.4
Nov 10, 16:04 UTC$0.0699$1.4
Nov 10, 04:05 UTC$0.0699$1.4
Nov 9, 16:04 UTC$0.0699$1.4
Nov 9, 04:04 UTC$0.0699$1.4
Nov 8, 16:04 UTC$0.0699$1.4
Nov 8, 04:03 UTC$0.0699$1.4
Nov 7, 16:04 UTC$0.0699$1.4
Nov 7, 04:04 UTC$0.0699$1.4
Nov 6, 16:04 UTC$0.0699$1.4
Nov 6, 04:04 UTC$0.0699$1.4
Nov 5, 16:05 UTC$0.0699$1.4
Nov 5, 10:05 UTC$0.0699$1.4
Nov 4, 16:02 UTC$0.08$3.6
Nov 4, 04:04 UTC$0.08$3.6
Nov 3, 16:04 UTC$0.08$3.6
Nov 3, 13:26 UTC$0.08$3.6
Nov 2, 16:01 UTC$0.08$3.6
Nov 2, 04:04 UTC$0.08$3.6
Nov 1, 16:03 UTC$0.08$3.6
Nov 1, 04:04 UTC$0.0699$1.4
Oct 31, 16:04 UTC$0.0699$1.4
Oct 31, 04:04 UTC$0.0699$1.4
Oct 30, 16:03 UTC$0.08$3.6
Oct 30, 04:04 UTC$0.08$3.6
Oct 29, 16:03 UTC$0.0699$1.4
Oct 29, 04:04 UTC$0.0699$1.4
Oct 28, 16:04 UTC$0.0699$1.4
Oct 28, 09:22 UTC$0.0699$1.4
Oct 27, 16:03 UTC$0.0699$1.4
Oct 27, 04:03 UTC$0.0769$1.4
Oct 26, 16:04 UTC$0.0699$1.4
Oct 26, 04:03 UTC$0.0699$1.4
Oct 25, 16:04 UTC$0.079$1.4
Oct 25, 06:03 UTC$0.079$1.4
Oct 24, 15:04 UTC$0.079$1.4
Oct 24, 03:03 UTC$0.079$1.4
Oct 23, 15:03 UTC$0.079$1.4
Oct 23, 03:04 UTC$0.079$1.4
Oct 22, 15:03 UTC$0.7$2.2
Oct 22, 05:04 UTC$0.079$1.4
Oct 21, 15:03 UTC$0.079$1.4
Oct 21, 03:02 UTC$0.079$1.4
Oct 20, 15:04 UTC$0.079$1.4
Oct 20, 03:03 UTC$0.079$1.4
Oct 19, 15:04 UTC$0.079$1.4
Oct 19, 10:50 UTC$0.079$1.4
Oct 18, 15:03 UTC$0.079$1.8
Oct 18, 03:05 UTC$0.079$4.8
Oct 17, 15:03 UTC$0.079$4.8
Oct 17, 03:03 UTC$0.079$4.8
Oct 16, 15:02 UTC$0.079$4.8
Oct 16, 03:03 UTC$0.079$4.8
Oct 15, 17:02 UTC$0.1$4.8
Oct 15, 03:02 UTC$0.079$4.8
Oct 14, 15:03 UTC$0.079$4.8
Oct 14, 03:02 UTC$0.079$4.8
Oct 13, 15:03 UTC$0.079$4.8
Oct 13, 05:03 UTC$0.079$4.8
Oct 12, 15:03 UTC$0.079$4.8
Oct 12, 03:04 UTC$0.079$4.8
Oct 11, 15:02 UTC$0.07$4.8
Oct 11, 03:03 UTC$0.039$4.8
Oct 10, 15:01 UTC$0.039$4.8
Oct 10, 03:04 UTC$0.039$4.8
Oct 9, 15:05 UTC$0.04$3.6
Oct 9, 03:04 UTC$0.039$3.39
Oct 8, 15:03 UTC$0.069$3.39
Oct 8, 03:05 UTC$0.049$3.39
Oct 7, 15:03 UTC$0.07$7
Oct 7, 03:03 UTC$0.07$7
Oct 6, 15:03 UTC$0.07$7
Oct 6, 03:02 UTC$0.07$7
Oct 5, 15:02 UTC$0.05$7
Oct 5, 03:04 UTC$1.4$4.4

Source: OpenRouter, as of Nov 20, 06:04 UTC

In the Drops

Models

Cloudflare: a GLM-5.3 serving bug put other users' data in 155 Workers AI responses

Cloudflare says prompts over 681,280 tokens hit an SGLang bug that reused stale KV cache, Aug 27 to Sep 9. It found no sign of deliberate exploitation and urges SGLang v0.5.22 or later.

Why it matters: Anyone self-hosting SGLang should check KV pool size against context length. Cloudflare's tests saw HTTP 200s and never checked the output.

From Drop #049 · Tue, Oct 27, 2026 · Morning
Models

Anthropic: open-weight GLM-5.3 builds exploits, and simple tricks beat its safeguards

Anthropic says GLM-5.3 built end-to-end V8 exploits in 50 of 410 attempts, near Claude Mythos Preview's 56. A red-team framing got it to engage 64% of the time, thinking prefill 92%, abliteration 100%.

Why it matters: Anthropic says anyone can download GLM-5.3, unlike similarly capable models released with safeguards or limited access, and urges governments to safety-test its successors.

From Drop #039 · Thu, Oct 22, 2026 · Morning
Models

AWS adds Z.ai's GLM 5.3 to Amazon Bedrock for eligible enterprise customers

AWS says the 753B-parameter mixture-of-experts model targets coding and long agent runs, with prompt caching, cross-Region inference and service tiers. AWS says Z.ai measured 84.5 on CyberGym.

Why it matters: AWS pairs it with Strix, an open-source pentesting agent, for testing apps you own, with inference running under your AWS account's controls.

From Drop #007 · Tue, Oct 6, 2026 · Morning