Dataset Viewer
Auto-converted to Parquet Duplicate
model_name
stringlengths
11
28
size
stringlengths
2
6
input_price_per_1m
float64
0.05
1.25
output_price_per_1m
float64
0.1
1.7
service
stringclasses
2 values
api_model_id
stringlengths
18
49
description
stringlengths
225
336
qwen2.5-7b-instruct
7B
0.2
0.2
NVIDIA
qwen/qwen2.5-7b-instruct
Qwen2.5-7B-Instruct represents an upgraded version of the Qwen model series, featuring significantly enhanced multilingual capabilities across diverse language tasks. This improved model is competitively priced at $0.20 per million input tokens and $0.20 per million output tokens.
gemma-2-9b-it
9B
0.1
0.1
NVIDIA
google/gemma-2-9b-it
Gemma-2-9B-IT is Google's 9-billion parameter instruction-tuned model from the Gemma 2 family, optimized for helpful and safe conversational interactions. This efficient model delivers strong performance at an exceptionally low price of $0.10 per million input tokens and $0.10 per million output tokens.
llama-3-8b-instruct-lite
8B
0.1
0.1
Together
meta-llama/Meta-Llama-3-8B-Instruct-Lite
Meta-Llama-3-8B-Instruct-Lite is Meta's 8-billion parameter model with INT4 quantization for lightweight and fast inference. This efficient model delivers strong instruction-following capabilities at an ultra-low price of $0.10 per million input and output tokens.
qwen2.5-7b-instruct-turbo
7B
0.3
0.3
Together
Qwen/Qwen2.5-7B-Instruct-Turbo
Qwen2.5-7B-Instruct-Turbo is Alibaba's 7-billion parameter model with turbo-optimized inference on Together AI, delivering fast and reliable multilingual performance. This model is priced at $0.30 per million input and output tokens.
mistral-7b-instruct-v0.3
7B
0.2
0.2
NVIDIA
mistralai/mistral-7b-instruct-v0.3
Mistral-7B-Instruct-v0.3 is a fast and efficient 7-billion parameter model specifically designed for instruction-following tasks. This streamlined model provides quick response times and reliable performance at cost-effective pricing of $0.20 per million input and output tokens.
qwen3-next-80b-a3b-instruct
80B
0.15
1.5
Together
Qwen/Qwen3-Next-80B-A3B-Instruct
Qwen3-Next-80B-A3B-Instruct is Alibaba's 80-billion parameter model with only 3-billion active parameters, achieving excellent efficiency through sparse activation. This model delivers strong performance at $0.15 per million input tokens and $1.50 per million output tokens.
llama3-70b-instruct
70B
0.9
0.9
NVIDIA
meta/llama3-70b-instruct
Llama3-70B-Instruct represents Meta's powerful 70-billion parameter model, maintaining robust capabilities and performance characteristics across a wide range of language tasks. This model provides comprehensive language understanding and generation at $0.90 per million input and output tokens.
mixtral-8x7b-instruct-v0.1
46.7B
0.6
0.6
NVIDIA
mistralai/mixtral-8x7b-instruct-v0.1
Mixtral-8x7B-Instruct-v0.1 is a 46.7-billion parameter Mixture of Experts (MoE) model composed of eight 7-billion parameter expert models, specifically optimized for creative text generation. This innovative architecture provides high-quality outputs while maintaining efficiency, available at $0.60 per million input an...
mixtral-8x22b-instruct-v0.1
140.6B
1.2
1.2
NVIDIA
mistralai/mixtral-8x22b-instruct-v0.1
Mixtral-8x22B-Instruct-v0.1 is a 140.6-billion parameter Mixture of Experts model comprising eight 22-billion parameter expert components. This large-scale MoE architecture delivers exceptional performance across diverse tasks while maintaining computational efficiency, priced at $1.20 per million input and output toke...
gpt-oss-20b
20B
0.05
0.2
Together
openai/gpt-oss-20b
GPT-OSS-20B is OpenAI's open-source 20-billion parameter model offering strong language capabilities at remarkably low cost. This model provides excellent value for general-purpose tasks, priced at $0.05 per million input tokens and $0.20 per million output tokens.
mistral-small-3-24b-instruct
24B
0.1
0.3
Together
mistralai/Mistral-Small-24B-Instruct-2501
Mistral-Small-3-24B-Instruct is Mistral AI's 24-billion parameter model designed for efficient instruction following with a balance of capability and speed. This mid-sized model handles a wide range of tasks reliably, priced at $0.10 per million input tokens and $0.30 per million output tokens.
llama-4-maverick
402B
0.27
0.85
Together
meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8
Llama-4-Maverick is Meta's cutting-edge Mixture of Experts model with 17-billion parameter experts across 128 expert slots, delivering frontier-level performance. This innovative architecture supports a massive 1M token context window, priced at $0.27 per million input tokens and $0.85 per million output tokens.
rnj-1-instruct
15B
0.15
0.15
Together
essentialai/rnj-1-instruct
RNJ-1-Instruct is Essential AI's efficient instruction-tuned model delivering strong reasoning and language capabilities. This model offers exceptional value at an ultra-low price of $0.15 per million input and output tokens.
gpt-oss-120b
120B
0.15
0.6
Together
openai/gpt-oss-120b
GPT-OSS-120B is OpenAI's open-source 120-billion parameter model delivering high-end performance across demanding language tasks. This large-scale model provides near-frontier capabilities at a competitive price of $0.15 per million input tokens and $0.60 per million output tokens.
qwen3-coder-next
200B
0.5
1.2
Together
Qwen/Qwen3-Coder-Next-FP8
Qwen3-Coder-Next is Alibaba's advanced code-specialized model optimized for programming tasks, code generation, and technical reasoning. This model delivers high-quality outputs at $0.50 per million input tokens and $1.20 per million output tokens.
deepseek-v3.1
671B
0.6
1.7
Together
deepseek-ai/DeepSeek-V3.1
DeepSeek-V3.1 is a 671-billion parameter Mixture of Experts model delivering state-of-the-art performance across coding, math, and reasoning tasks. This powerful model provides exceptional capabilities at $0.60 per million input tokens and $1.70 per million output tokens.
llama-3.3-70b-instruct-turbo
70B
0.88
0.88
Together
meta-llama/Llama-3.3-70B-Instruct-Turbo
Llama-3.3-70B-Instruct-Turbo is Meta's 70-billion parameter model with turbo-optimized inference, providing fast and high-quality responses for complex tasks. This model supports a 131K token context window, priced at $0.88 per million input and output tokens.
cogito-v2-1-671b
671B
1.25
1.25
Together
deepcogito/cogito-v2-1-671b
Cogito-V2.1-671B is Deep Cogito's 671-billion parameter model delivering powerful reasoning and language understanding comparable to frontier models. This large-scale model provides exceptional performance at $1.25 per million input and output tokens, offering a cost-effective alternative for complex reasoning tasks.

xRouteBench — LLM Routing Benchmark

xRouteBench is a benchmark for training and evaluating LLM routers — systems that pick the best LLM from a candidate pool for each incoming query, trading off performance vs. price cost.

Every query in each scenario was executed against all 18 candidate LLMs, recording each model's response, task performance, token usage, and latency. A router learns from the train split which model to pick, and is evaluated on test.

Scenarios (configs)

Config Domain Train rows / queries Test rows / queries Metric
llmrouter_generic 13 classic NLP benchmarks (MMLU, GSM8K, MATH, MBPP, ARC, …) 80,802 / 4,487 67,122 / 3,729 em_mc, GSM8K, MATH, code_eval, f1
memory_locomo Long-conversation memory QA (RAG top-k=5) 15,930 / 885 5,652 / 314 f1
memory_longmemeval Long-term memory eval (RAG top-k=5) 4,986 / 277 1,818 / 101 f1
timeseries Time-series understanding (7 sub-tasks) 17,568 / 976 2,286 / 127 mc
video Egocentric video QA (Charades-Ego) 3,618 / 201 486 / 27 em
multimodal_geometry3k Geometry math (multimodal) 8,640 / 480 1,098 / 61 em
multimodal_mathvista Visual math reasoning 14,400 / 800 1,800 / 100 em, em_mc
personalized Personalized preference (LLM-judge; chat-format queries) 2,464 / 2,235 308 / 303 llm_judge
llm_candidates The 18-model candidate pool with pricing 18 models

Schema (routing data)

Each row = one (query, candidate model) pair:

Field Type Description
task_name str Sub-task the query belongs to (e.g. gsm8k, mbpp)
query str Full input prompt (personalized: JSON-encoded chat messages)
ground_truth list/str Reference answer(s)
metric str Scoring metric for this row
choices str Options for multiple-choice items (JSON-encoded)
task_id str ID within the sub-task
model_name str Candidate LLM this row was executed with
response str The model's actual response
token_num int Total tokens
input_tokens / output_tokens int Token breakdown (for price computation)
response_time float Latency in seconds
performance float Task score of this model on this query (0–1)
embedding_id int Index into precomputed query-embedding files (not included here)

Raw query configs (*_queries)

Each scenario also ships its raw query set (no model executions) as a <scenario>_queries config with train / valid / test splits — use these to run your own candidate models from scratch. Fields: task_name, query, ground_truth, metric, choices, task_id (+ conversation_id/category for the memory scenarios). The memory queries are the RAG top-k=5 turn-pair variant used in the published experiments. Note: the valid split exists only here; the routing-data configs have train/test.

Candidate pool & pricing (llm_candidates config)

18 models spanning $0.05–$1.25 per 1M input tokens (25× spread) served via Together AI / NVIDIA NIM. Row cost = input_tokens × input_price/1e6 + output_tokens × output_price/1e6.

Usage

from datasets import load_dataset

ds = load_dataset("ulab-ai/xRouteBench", "llmrouter_generic")   # any config name above
train, test = ds["train"], ds["test"]

pricing = load_dataset("ulab-ai/xRouteBench", "llm_candidates")["train"]

Composite reward for cost-aware routing (GraphRouter-style):

reward = α · norm(performance) − β · norm(price_cost)

Notes

  • Query embeddings (.pt) are not included; they can be regenerated from the query field with any sentence encoder.
  • The memory scenarios use the RAG top-k=5 turn-pair context variant.
Downloads last month
28