z-ai

GLM 5.3 Flash

textimagevideo

Independent comparisons

No exact benchmark result or ECI estimate is linked to this model. Another version’s score is not substituted.

Choose models to compare

What changes your bill?

Illustrative token baskets, not measured tasks.

Loading chart

Short exchange1K input + 250 output tokens$0.0003

Long answer4K input + 4K output tokens$0.0026

Document analysis100K input + 2K output tokens$0.016

Repeated workload1M input + 250K output tokens$0.275

List prices, captured 2026-10-25. Excludes caching, batch discounts, tools and retries. Token baskets are not estimates of a specific task’s token use.

Within the same developer

Five lowest listed workload costs, plus this model.

Loading chart

Fixed basket: 1M input + 250k output. Logarithmic positions; labels show actual USD. Lower cost does not imply equivalent capability.

PUBLISHED LIMITS

Room to read
Room to respond

Context and maximum output are separate limits, not a quality score. They are not added together.

Context window1Mtokens

Maximum output943.7Ktokens

No independent evaluation is linked to this exact model. The charts above compare published prices and specifications.

Published specifications

Context window1Mtokens

API input price$0.15USD / million tokens

API output price$0.5USD / million tokens

Maximum output943.7Kmaximum tokens per response

Output modalitiestextpublished output types

Supported parameters21API parameters listed

Model information

Provider: z-ai
Input modalities: text, image, video
Output modalities: text
Maximum output: 943,717 tokens
Supported parameters: frequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, parallel_tool_calls, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p
Added to catalogue: 2026-08-26
Knowledge cutoff: Not available
Catalogue snapshot: 2026-10-25
Source: OpenRouter Models API ↗

Prices are per million tokens. The workload estimate uses 1,000 input and 250 output tokens. Context describes request capacity, not model quality.