Datasets:
File size: 3,828 Bytes
a6c60d7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 | {
"note": "Registry of result labels. 'order' of this list fixes the bootstrap RNG stream: append new models at the end so existing confidence intervals never move.",
"models": [
{
"label": "alibaba/qwen-flash",
"model_ref": "qwen-flash",
"route": "Alibaba Cloud Model Studio API, buffered (non-streaming), enable_thinking=false, temperature 0",
"hardware_class": "hosted API; serving hardware not disclosed",
"output_token_cap": {"R1-R5": 512, "R6-R7": 2048},
"latency_instrument": "Buffered API: wall time only; TTFT and tok/s unavailable"
},
{
"label": "alibaba/qwen-plus",
"model_ref": "qwen-plus",
"route": "Alibaba Cloud Model Studio API, buffered (non-streaming), enable_thinking=false, temperature 0",
"hardware_class": "hosted API; serving hardware not disclosed",
"output_token_cap": {"R1-R5": 512, "R6-R7": 2048},
"latency_instrument": "Buffered API: wall time only; TTFT and tok/s unavailable",
"coverage_note": "Run stopped during R6; R6 partial and R7 not run. Missing trials are not failures."
},
{
"label": "RK1-Qwen3-4B",
"model_ref": "Qwen3-4B-Instruct-2507, RKLLM W8A8 build, compiled context 2048",
"route": "owned edge node, local NPU inference via the RKLLM runtime (1.3.1)",
"hardware_class": "edge SoC: Turing RK1 compute module, Rockchip RK3588, 3-core NPU, 32 GB",
"output_token_cap": {"R1-R5": 512, "R6-R7": 512},
"latency_instrument": "RKLLM native callbacks/perf; trial wall time includes remote shell and model init",
"coverage_note": "Native 512 output / 2048 context limit kept for R6-R7, unlike the 2048 output cap on hosted routes. Input tokens were not logged."
},
{
"label": "astra",
"model_ref": "openai/gpt-6-astra",
"route": "OpenRouter, OpenAI-compatible streaming chat completions, temperature 0",
"hardware_class": "hosted API; serving hardware not disclosed",
"output_token_cap": {"R1-R5": 512, "R6-R7": 2048},
"latency_instrument": "SSE first content event; tok/s = (completion_tokens-1)/(last-first content event)"
},
{
"label": "deepseek-v4-flash",
"model_ref": "deepseek/deepseek-v4-flash",
"route": "OpenRouter, OpenAI-compatible streaming chat completions, temperature 0",
"hardware_class": "hosted API; serving hardware not disclosed",
"output_token_cap": {"R1-R5": 512, "R6-R7": 2048},
"latency_instrument": "SSE first content event; tok/s = (completion_tokens-1)/(last-first content event)"
},
{
"label": "glm-4.7-flash",
"model_ref": "z-ai/glm-4.7-flash",
"route": "OpenRouter, OpenAI-compatible streaming chat completions, temperature 0",
"hardware_class": "hosted API; serving hardware not disclosed",
"output_token_cap": {"R1-R5": 512, "R6-R7": 2048},
"latency_instrument": "SSE first content event; tok/s = (completion_tokens-1)/(last-first content event)"
},
{
"label": "gpt-5-nano",
"model_ref": "openai/gpt-5-nano",
"route": "OpenRouter, OpenAI-compatible streaming chat completions, temperature 0",
"hardware_class": "hosted API; serving hardware not disclosed",
"output_token_cap": {"R1-R5": 512, "R6-R7": 2048},
"latency_instrument": "SSE first content event; tok/s = (completion_tokens-1)/(last-first content event)"
},
{
"label": "qwen3.7-flash",
"model_ref": "qwen/qwen3.7-flash",
"route": "OpenRouter, OpenAI-compatible streaming chat completions, temperature 0",
"hardware_class": "hosted API; serving hardware not disclosed",
"output_token_cap": {"R1-R5": 512, "R6-R7": 2048},
"latency_instrument": "SSE first content event; tok/s = (completion_tokens-1)/(last-first content event)"
}
]
}
|