Output limits
The longest single response each of 38 tracked models can return, largest first, with the equivalent word count and what a full-length answer costs at list price.
Limits from the AI Gateway catalog on September 3, 2026; prices are the live list price. Words assume 0.75 words per token.
| Model | Provider | Max output | ~Words | Context | Output $/M | Full response |
|---|---|---|---|---|---|---|
| GLM-5.3 | Z.ai | 1M tokens | ~750,000 | 1M tokens | $4.40 | $4.40 |
| Grok 4.5 | xAI | 500k tokens | ~375,000 | 500k tokens | $6.00 | $3.00 |
| Grok 4.6 | xAI | 500k tokens | ~375,000 | 500k tokens | $6.00 | $3.00 |
| DeepSeek V4-Flash | DeepSeek | 384k tokens | ~288,000 | 1M tokens | $0.28 | $0.11 |
| DeepSeek V4-Pro | DeepSeek | 384k tokens | ~288,000 | 1M tokens | $0.87 | $0.33 |
| Mistral Medium 3.5 | Mistral | 256k tokens | ~192,000 | 256k tokens | $7.50 | $1.92 |
| Muse Glimmer 30B | Meta | 131k tokens | ~98,304 | 131k tokens | $1.50 | $0.20 |
| Kimi K3 | Moonshot | 131k tokens | ~98,304 | 1M tokens | $15.00 | $1.97 |
| GLM-5.3 Flash | Z.ai | 131k tokens | ~98,250 | 1M tokens | $0.50 | $0.066 |
| Qwen 3.8 Flash | Alibaba | 128k tokens | ~96,000 | 991k tokens | $0.47 | $0.060 |
| GPT-5.6 Luna | OpenAI | 128k tokens | ~96,000 | 1.1M tokens | $1.20 | $0.15 |
| GLM-5.2 | Z.ai | 128k tokens | ~96,000 | 1M tokens | $4.40 | $0.56 |
| Qwen 3.8 Max | Alibaba | 128k tokens | ~96,000 | 1M tokens | $6.00 | $0.77 |
| Claude Sonnet 5 | Anthropic | 128k tokens | ~96,000 | 1M tokens | $10.00 | $1.28 |
| GPT-5.6 Sol | OpenAI | 128k tokens | ~96,000 | 1.1M tokens | $10.00 | $1.28 |
| GPT-5.6 Terra | OpenAI | 128k tokens | ~96,000 | 1.1M tokens | $12.00 | $1.54 |
| GPT-5.2 | OpenAI | 128k tokens | ~96,000 | 400k tokens | $14.00 | $1.79 |
| GPT-5.4 | OpenAI | 128k tokens | ~96,000 | 1.1M tokens | $15.00 | $1.92 |
| Claude Sonnet 4.6 | Anthropic | 128k tokens | ~96,000 | 1M tokens | $15.00 | $1.92 |
| Claude Opus 4.6 | Anthropic | 128k tokens | ~96,000 | 1M tokens | $25.00 | $3.20 |
| Claude Opus 4.8 | Anthropic | 128k tokens | ~96,000 | 1M tokens | $25.00 | $3.20 |
| Claude Opus 4.7 | Anthropic | 128k tokens | ~96,000 | 1M tokens | $25.00 | $3.20 |
| Claude Opus 5 | Anthropic | 128k tokens | ~96,000 | 1M tokens | $25.00 | $3.20 |
| GPT-5.5 | OpenAI | 128k tokens | ~96,000 | 1M tokens | $30.00 | $3.84 |
| Claude Fable 5.1 | Anthropic | 128k tokens | ~96,000 | 1M tokens | $50.00 | $6.40 |
| Claude Fable 5 | Anthropic | 128k tokens | ~96,000 | 1M tokens | $50.00 | $6.40 |
| Gemini 3.7 Flash | 66k tokens | ~49,152 | 1M tokens | $3.75 | $0.25 | |
| Gemini 3.5 Flash-Lite | 65k tokens | ~48,750 | 1M tokens | $2.50 | $0.16 | |
| Gemini 3.6 Flash | 64k tokens | ~48,000 | 1M tokens | $3.75 | $0.24 | |
| Qwen 3.7 Max | Alibaba | 64k tokens | ~48,000 | 991k tokens | $7.50 | $0.48 |
| Gemini 3.5 Flash | 64k tokens | ~48,000 | 1M tokens | $9.00 | $0.58 | |
| Gemini 3.1 Pro | 64k tokens | ~48,000 | 1M tokens | $12.00 | $0.77 | |
| Kimi K2.7 Code | Moonshot | 33k tokens | ~24,576 | 256k tokens | $4.00 | $0.13 |
| GPT-4o Mini | OpenAI | 16k tokens | ~12,288 | 128k tokens | $0.60 | $0.010 |
| GPT-4o | OpenAI | 16k tokens | ~12,288 | 128k tokens | $10.00 | $0.16 |
| Llama 4 Scout | Meta | 8k tokens | ~6,144 | 128k tokens | $1.20 | $0.010 |
| Llama 4 Maverick | Meta | 8k tokens | ~6,144 | 128k tokens | $1.60 | $0.013 |
| DeepSeek R1 | DeepSeek | 8k tokens | ~6,144 | 128k tokens | $2.19 | $0.018 |
Full response = max output tokens × output price, a ceiling at list rates. Models without a published output limit are not listed.
It is the longest single response a model will return in one request, measured in tokens (about 0.75 English words each). It is separate from the context window: the context window is how much the model can read, including your prompt; the max output is how much it can write back at once. If you need more than the limit, you continue in a follow-up turn.
GLM-5.3 (Z.ai) has the largest maximum output of the 38 models tracked here: 1M tokens per response, roughly 750,000 words. 26 models can return 128K tokens or more in one response.
Among models that can return at least 64K tokens in one response, DeepSeek V4-Flash has the lowest output price at $0.28/1M tokens. A maximum-length 384k tokens response costs about $0.11 at list price. Long outputs are billed on output tokens, which are usually several times the input rate.
Output tokens × output price. For GLM-5.3, a full 1M tokens response costs about $4.40 at $4.4/1M. The table above shows the same figure for every model; it is a ceiling, since most responses stop well short of the limit.
Models end a response when they judge it complete, when your request sets a lower limit (max_tokens), or when the context window fills up because a long prompt plus a long answer must fit together. The published maximum is what the API allows, not what the model usually produces. If you need a long document, ask for it in sections.
DeepSeek R1 has the smallest published maximum on this page at 8k tokens per response. Smaller limits are common on older or speed-focused models and are fine for chat, classification, and short drafts.