languagebench / models.json
davidpomerenke's picture
Upload from GitHub Actions: ran full evaluation locally
088f96f verified
raw
history blame
20.9 kB
[
{
"id":"ai21\/jamba-large-1.7",
"name":"Jamba Large 1.7",
"provider_name":"AI21",
"cost":8.0,
"hf_id":"ai21labs\/AI21-Jamba-Large-1.7",
"size":398555145696.0,
"type":"open-source",
"license":"Other",
"creation_date":1751414400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"amazon\/nova-micro-v1",
"name":"Nova Micro 1.0",
"provider_name":"Amazon",
"cost":0.14,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1733356800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"anthracite-org\/magnum-v4-72b",
"name":"Magnum v4 72B",
"provider_name":"Magnum v4 72B",
"cost":5.0,
"hf_id":"anthracite-org\/magnum-v4-72b",
"size":72706203648.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1726790400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"anthropic\/claude-3.7-sonnet",
"name":"Claude 3.7 Sonnet",
"provider_name":"Anthropic",
"cost":15.0,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1740355200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"anthropic\/claude-sonnet-4",
"name":"Claude Sonnet 4",
"provider_name":"Anthropic",
"cost":15.0,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1747872000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"arcee-ai\/virtuoso-large",
"name":"Virtuoso Large",
"provider_name":"Arcee AI",
"cost":1.2,
"hf_id":"arcee-ai\/Virtuoso-Large",
"size":72706203648.0,
"type":"open-source",
"license":"Other",
"creation_date":1749427200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"arliai\/qwq-32b-arliai-rpr-v1",
"name":"QwQ 32B RpR v1",
"provider_name":"ArliAI",
"cost":0.0,
"hf_id":"ArliAI\/QwQ-32B-ArliAI-RpR-v1",
"size":32763876352.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1743984000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"baidu\/ernie-4.5-300b-a47b",
"name":"ERNIE 4.5 300B A47B ",
"provider_name":"Baidu",
"cost":1.1,
"hf_id":"baidu\/ERNIE-4.5-300B-A47B-PT",
"size":300474051776.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1751068800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"deepseek\/deepseek-chat",
"name":"DeepSeek V3",
"provider_name":"DeepSeek",
"cost":0.8,
"hf_id":"deepseek-ai\/DeepSeek-V3",
"size":684531386000.0,
"type":"open-source",
"license":"",
"creation_date":1735084800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"deepseek\/deepseek-chat-v3-0324",
"name":"DeepSeek V3 0324",
"provider_name":"DeepSeek",
"cost":0.0,
"hf_id":"deepseek-ai\/DeepSeek-V3-0324",
"size":684531386000.0,
"type":"open-source",
"license":"Mit",
"creation_date":1742774400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"deepseek\/deepseek-chat-v3.1",
"name":"DeepSeek V3.1",
"provider_name":"DeepSeek",
"cost":0.0,
"hf_id":"deepseek-ai\/DeepSeek-V3.1",
"size":684531386000.0,
"type":"open-source",
"license":"Mit",
"creation_date":1755734400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"eleutherai\/llemma_7b",
"name":"Llemma 7b",
"provider_name":"EleutherAI",
"cost":1.2,
"hf_id":"EleutherAI\/llemma_7b",
"size":null,
"type":"open-source",
"license":"Llama2",
"creation_date":1694476800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"google\/gemini-2.0-flash-001",
"name":"Gemini 2.0 Flash",
"provider_name":"Google",
"cost":0.4,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1738713600000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"google\/gemini-2.0-flash-lite-001",
"name":"Gemini 2.0 Flash Lite",
"provider_name":"Google",
"cost":0.3,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1740441600000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"google\/gemini-2.5-flash",
"name":"Gemini 2.5 Flash",
"provider_name":"Google",
"cost":2.5,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1750118400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"google\/gemma-2-9b-it",
"name":"Gemma 2 9B",
"provider_name":"Google",
"cost":0.0,
"hf_id":"google\/gemma-2-9b-it",
"size":9241705984.0,
"type":"open-source",
"license":"Gemma",
"creation_date":1719187200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"google\/gemma-3-12b-it",
"name":"Gemma 3 12B",
"provider_name":"Google",
"cost":0.0,
"hf_id":"google\/gemma-3-12b-it",
"size":12187325040.0,
"type":"open-source",
"license":"Gemma",
"creation_date":1740787200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"google\/gemma-3-27b-it",
"name":"Gemma 3 27B",
"provider_name":"Google",
"cost":0.0,
"hf_id":"google\/gemma-3-27b-it",
"size":27432406640.0,
"type":"open-source",
"license":"Gemma",
"creation_date":1740787200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"google\/gemma-3-4b-it",
"name":"Gemma 3 4B",
"provider_name":"Google",
"cost":0.0,
"hf_id":"google\/gemma-3-4b-it",
"size":4300079472.0,
"type":"open-source",
"license":"Gemma",
"creation_date":1740009600000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"liquid\/lfm-7b",
"name":"LFM 7B",
"provider_name":"Liquid",
"cost":0.01,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1737763200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"meta-llama\/llama-3-70b-instruct",
"name":"Llama 3 70B Instruct",
"provider_name":"Meta",
"cost":0.4,
"hf_id":"meta-llama\/Meta-Llama-3-70B-Instruct",
"size":70553706496.0,
"type":"open-source",
"license":"Llama3",
"creation_date":1713312000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"meta-llama\/llama-3.1-70b-instruct",
"name":"Llama 3.1 70B Instruct",
"provider_name":"Meta",
"cost":0.28,
"hf_id":"meta-llama\/Llama-3.1-70B-Instruct",
"size":70553706496.0,
"type":"open-source",
"license":"Llama3.1",
"creation_date":1721088000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"meta-llama\/llama-3.3-70b-instruct",
"name":"Llama 3.3 70B Instruct",
"provider_name":"Meta",
"cost":0.0,
"hf_id":"meta-llama\/Llama-3.3-70B-Instruct",
"size":70553706496.0,
"type":"open-source",
"license":"Llama3.3",
"creation_date":1732579200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"meta-llama\/llama-4-maverick",
"name":"Llama 4 Maverick",
"provider_name":"Meta",
"cost":0.0,
"hf_id":"meta-llama\/Llama-4-Maverick-17B-128E-Instruct",
"size":401583781376.0,
"type":"open-source",
"license":"Other",
"creation_date":1743465600000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"microsoft\/phi-4",
"name":"Phi 4",
"provider_name":"Microsoft",
"cost":0.14,
"hf_id":"microsoft\/phi-4",
"size":14659507200.0,
"type":"open-source",
"license":"Mit",
"creation_date":1733875200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"microsoft\/phi-4-multimodal-instruct",
"name":"Phi 4 Multimodal Instruct",
"provider_name":"Microsoft",
"cost":0.1,
"hf_id":"microsoft\/Phi-4-multimodal-instruct",
"size":5574460384.0,
"type":"open-source",
"license":"Mit",
"creation_date":1740355200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"mistralai\/mistral-nemo",
"name":"Mistral Nemo",
"provider_name":"Mistral",
"cost":0.0,
"hf_id":"mistralai\/Mistral-Nemo-Instruct-2407",
"size":12247782400.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1721174400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"mistralai\/mistral-saba",
"name":"Saba",
"provider_name":"Mistral",
"cost":0.6,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1739750400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"mistralai\/mistral-small-3.1-24b-instruct",
"name":"Mistral Small 3.1 24B",
"provider_name":"Mistral",
"cost":0.0,
"hf_id":"mistralai\/Mistral-Small-3.1-24B-Instruct-2503",
"size":24011361280.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1741651200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"mistralai\/mixtral-8x7b-instruct",
"name":"Mixtral 8x7B Instruct",
"provider_name":"Mistral",
"cost":0.24,
"hf_id":"mistralai\/Mixtral-8x7B-Instruct-v0.1",
"size":46702792704.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1702166400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"moonshotai\/kimi-k2",
"name":"Kimi K2",
"provider_name":"MoonshotAI",
"cost":0.0,
"hf_id":"moonshotai\/Kimi-K2-Instruct",
"size":null,
"type":"open-source",
"license":"Other",
"creation_date":1752192000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"nousresearch\/deephermes-3-llama-3-8b-preview",
"name":"DeepHermes 3 Llama 3 8B Preview",
"provider_name":"Nous",
"cost":0.0,
"hf_id":"NousResearch\/DeepHermes-3-Llama-3-8B-Preview",
"size":8030261248.0,
"type":"open-source",
"license":"Llama3",
"creation_date":1739318400000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"openai\/gpt-4.1",
"name":"GPT-4.1",
"provider_name":"OpenAI",
"cost":8.0,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1744588800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"openai\/gpt-4.1-mini",
"name":"GPT-4.1 Mini",
"provider_name":"OpenAI",
"cost":1.6,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1744588800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"openai\/gpt-4.1-nano",
"name":"GPT-4.1 Nano",
"provider_name":"OpenAI",
"cost":0.4,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1744588800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"openai\/gpt-4o-2024-11-20",
"name":"GPT-4o (2024-11-20)",
"provider_name":"OpenAI",
"cost":10.0,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1732060800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"openai\/gpt-4o-mini",
"name":"GPT-4o-mini",
"provider_name":"OpenAI",
"cost":0.6,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1721260800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"openai\/gpt-5",
"name":"GPT-5",
"provider_name":"OpenAI",
"cost":10.0,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1754524800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"openai\/gpt-5-nano",
"name":"GPT-5 Nano",
"provider_name":"OpenAI",
"cost":0.4,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1754524800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"openai\/gpt-oss-120b",
"name":"gpt-oss-120b",
"provider_name":"OpenAI",
"cost":0.0,
"hf_id":"openai\/gpt-oss-120b",
"size":120412337472.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1754265600000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"qwen\/qwen-2.5-72b-instruct",
"name":"Qwen2.5 72B Instruct",
"provider_name":"Qwen2.5 72B Instruct (free)",
"cost":0.0,
"hf_id":"Qwen\/Qwen2.5-72B-Instruct",
"size":72706203648.0,
"type":"open-source",
"license":"Other",
"creation_date":1726444800000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"qwen\/qwen-2.5-coder-32b-instruct",
"name":"Qwen2.5 Coder 32B Instruct",
"provider_name":"Qwen2.5 Coder 32B Instruct (free)",
"cost":0.0,
"hf_id":"Qwen\/Qwen2.5-Coder-32B-Instruct",
"size":32763876352.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1730851200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"qwen\/qwen3-235b-a22b",
"name":"Qwen3 235B A22B",
"provider_name":"Qwen",
"cost":0.0,
"hf_id":"Qwen\/Qwen3-235B-A22B",
"size":235093634560.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1745712000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"qwen\/qwen3-30b-a3b",
"name":"Qwen3 30B A3B",
"provider_name":"Qwen",
"cost":0.0,
"hf_id":"Qwen\/Qwen3-30B-A3B",
"size":30532122624.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1745712000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"qwen\/qwen3-32b",
"name":"Qwen3 32B",
"provider_name":"Qwen",
"cost":0.07,
"hf_id":"Qwen\/Qwen3-32B",
"size":32762123264.0,
"type":"open-source",
"license":"Apache 2.0",
"creation_date":1745712000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"tencent\/hunyuan-a13b-instruct",
"name":"Hunyuan A13B Instruct",
"provider_name":"Tencent",
"cost":0.0,
"hf_id":"tencent\/Hunyuan-A13B-Instruct",
"size":80393183232.0,
"type":"open-source",
"license":"Other",
"creation_date":1750809600000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"undi95\/remm-slerp-l2-13b",
"name":"ReMM SLERP 13B",
"provider_name":"ReMM SLERP 13B",
"cost":0.65,
"hf_id":"Undi95\/ReMM-SLERP-L2-13B",
"size":null,
"type":"open-source",
"license":"Cc By Nc 4.0",
"creation_date":1693785600000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"x-ai\/grok-4",
"name":"Grok 4",
"provider_name":"xAI",
"cost":15.0,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":1752019200000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"z-ai\/glm-4.5v",
"name":"GLM 4.5V",
"provider_name":"Z.AI",
"cost":1.8,
"hf_id":"zai-org\/GLM-4.5V",
"size":107710933120.0,
"type":"open-source",
"license":"Mit",
"creation_date":1754784000000,
"tasks":[
"translation_from",
"translation_to",
"classification",
"mmlu",
"arc",
"truthfulqa",
"mgsm"
]
},
{
"id":"google\/translate-v2",
"name":"Google Translate",
"provider_name":"Google",
"cost":20.0,
"hf_id":null,
"size":null,
"type":"closed-source",
"license":null,
"creation_date":null,
"tasks":[
"translation_from",
"translation_to"
]
}
]