Start calling this model with endpoint-specific examples.
Headline benchmark standings and comparison context.
Key dates, capabilities, and model metadata.
Start calling this model with endpoint-specific examples.
Headline benchmark standings and comparison context.
Key dates, capabilities, and model metadata.
Start calling this model with endpoint-specific examples.
Headline benchmark standings and comparison context.
Key dates, capabilities, and model metadata.
26 Jan 2025
26 Jan 2025
Parameters
73,000,000,000
License
Apache 2.0
Input
Output
Core latency and throughput trends from recent traffic.
Headline benchmark standings and comparison context.
Top benchmark results for qwen/qwen2.5-vl-72b.
Detailed benchmark comparisons now live in the Compare tool.
Latency
Throughput
Uptime
Latency
Throughput
Uptime
Latency
Throughput
Uptime
Start calling this model with endpoint-specific examples.
Choose a supported endpoint, pick a main language, then select the example style you want to copy.
import AIStats from '@ai-stats/sdk';
const client = new AIStats({
apiKey: process.env.AI_STATS_API_KEY,
});
const response = await client.generateResponse({
"model": "qwen/qwen2.5-vl-72b",
"input": "Give me one fun fact about cURL.",
"service_tier": "standard"
});
const outputText = response.output
?.flatMap((item) => item.content ?? [])
.find((item) => item.type === "output_text")
?.text;
console.log(outputText ?? response);Parameters
Aggregated across active providers for the responses route.
Routing will select a compatible provider when a parameter narrows availability, so this list stays model-facing instead of provider-facing.
| Parameter | Description |
|---|---|
temperature | Controls how random token selection can be. |
top_p | Applies nucleus sampling by limiting candidates to a probability mass threshold. |
top_k | Restricts sampling to the top-k candidate tokens on providers that expose it. |
max_tokens | Caps output length on endpoints and providers that use the max_tokens field name. |
frequency_penalty | Discourages repeated tokens in proportion to how often they already appeared. |
presence_penalty | Encourages the model to explore new wording or topics after they first appear. |
repetition_penalty | Applies provider-specific anti-repetition behavior outside the classic penalty fields. |
seed | Requests deterministic sampling when the upstream provider supports seeded generation. |
stop | Defines one or more sequences that terminate generation early. |
logprobs | Requests token-level probability data in the response. |
logit_bias | Adjusts token selection bias directly when a provider exposes logit control. |
min_p | Narrows sampling by discarding tokens below a minimum probability threshold. |
top_logprobs | Limits how many alternative token probabilities are returned per position. |
Select a route to update the request snippet and compatibility details.
/v1/responses/v1/chat/completions/v1/messages