Urgent.News

What's breaking now, across thousands of outlets.

AI

GLM-5.3-Flash Intelligence, Performance and Price Analysis

GLM-5.3-Flash is a leading open weight model in intelligence, comparable to other models of its size but with slower performance and verbose outputs. It supports both text and image input, generating text, and boasts a 1M token context window. The model scores 57 on the Artificial Analysis Intelligence Index, outperforming the median of 27.

GLM-5.3-Flash generates 150 million tokens, significantly more verbose than the median of 110 million. The model is moderately priced at $0.15 per 1M input tokens and $0.50 per 1M output tokens, costing $138.02 to evaluate on the Intelligence Index. At 50 tokens per second, it lags behind the average speed of 66 tokens per second.

Metrics are compared against a set of models, including GDPval-AA v2, τ³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, and AA-LCR. The model is labeled as Non-commercial use, meaning commercial use is not restricted. It can handle agentic real-world tasks and has an AA-Omniscience Index score of 57, indicating a higher reliability in knowledge and fewer hallucinations.

The model's cost per task is calculated based on input, cache hit, cache write, reasoning, and answer token prices, with the weighted average cost per task being $0.38. The number of tokens required per task is determined by multiplying the output tokens by the benchmark weights, then dividing by task count. GLM-5.3-Flash's cache hit price is $0.025 per million tokens, offering a discount compared to regular input prices.

The model's maximum combined input and output tokens are 1M. It can process and generate approximately 50 tokens per second, with a time to first answer token of 6.6 seconds and 500 tokens received in 38.4 seconds. The model consists of 1 billion trainable weights and biases.

Written by urgent.news from Hacker News's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at artificialanalysis.ai →

More in AI

We need better than Pay-to-Crawl

The data-hunger of current AIs is reviving an interesting old idea: pay-to-use internet (pay-to-crawl in this case). This development could have very positive ramifications in principle, but current…

Ex-Meta scientists want to bring visual AI to the factory floor

Perceptron offers an AI model that it says can help machines navigate the world while also providing in-depth visual intelligence.

  • Former Meta scientists Armen Aghajanyan and Akshat Shrivastava founded Perceptron.
  • Isaac 0.5, Perceptron's latest model, assists vision-guided robots in industrial settings.
  • The startup raised $21 million in funding for its general-purpose visual AI software.

More from Wednesday 26 August →