Fournisseurs d'IA
Explorez la latence et les prix des principaux fournisseurs d'IA
Provider
Best Price
Best Latency
# of Models
Deepinfra
AI Provider
Best Price
66 Modèle Modèles
Best Latency
59 Modèle Modèles
# Models
77 Modèle Modèles
Model | Latency | Blended $ | Input $ | Output $ |
|---|---|---|---|---|
| llama-3-2-1b | 1st 0.44 | 2nd $0.01 | $0.01 | $0.01 |
| llama-3-1-8b-turbo-fp8 | 1st 0.18 | 1st $0.02 | $0.02 | $0.03 |
| llama-3-2-3b | 1st 0.46 | 1st $0.02 | $0.02 | $0.02 |
| mistral-nemo | 1st 0.19 | 1st $0.03 | $0.02 | $0.04 |
| mistral-7b | 3rd 0.45 | 1st $0.03 | $0.03 | $0.05 |
| llama-3-1-8b | 0.28 | 3rd $0.04 | $0.03 | $0.05 |
| llama-3-8b | 1st 0.31 | 1st $0.04 | $0.03 | $0.06 |
| llama-3-2-11b-vision | 2nd 1.92 | 1st $0.05 | $0.05 | $0.05 |
| gemma-3-4b | 1st 0.35 | 1st $0.05 | $0.04 | $0.08 |
| deepseek-ocr | 1st 0.17 | 2nd $0.05 | $0.03 | $0.10 |
| gpt-oss-20b-high | 2nd 0.18 | 2nd $0.06 | $0.03 | $0.14 |
| gemma-3-12b | 1st 0.29 | 1st $0.06 | $0.04 | $0.13 |
| mistral-small-3 | 3rd 0.39 | 1st $0.06 | $0.05 | $0.08 |
| mistral-small-3-1 | 3rd 0.39 | 1st $0.06 | $0.05 | $0.10 |
| nvidia-nemotron-nano-9b-v2 | 1st 0.42 | 1st $0.07 | $0.04 | $0.16 |
| devstral-small-may | 1st 0.25 | 1st $0.07 | $0.06 | $0.12 |
| gpt-oss-120b-high | 2nd 0.26 | 1st $0.08 | $0.04 | $0.19 |
| qwen2-5-coder-32b | 2nd 0.40 | 1st $0.08 | $0.06 | $0.15 |
| phi-4 | 1st 0.31 | 1st $0.09 | $0.07 | $0.14 |
| nvidia-nemotron-3-nano | 1st 0.24 | 1st $0.10 | $0.06 | $0.24 |
| mistral-small-3-2-fp8 | 1st 0.25 | 1st $0.11 | $0.07 | $0.20 |
| gemma-3-27b | 1st 0.36 | 1st $0.11 | $0.09 | $0.16 |
| qwen3-14b-fp8 | 1st 0.21 | 1st $0.12 | $0.08 | $0.24 |
| devstral-small | 1st 0.26 | 1st $0.12 | $0.07 | $0.28 |
| qwen3-coder-30b-a3b-fp8 | 1st 0.22 | 1st $0.12 | $0.07 | $0.26 |
| qwen3-30b-fp8 | 1st 0.21 | 1st $0.13 | $0.08 | $0.29 |
| llama-4-scout | 0.32 | 2nd $0.14 | $0.08 | $0.30 |
| qwq-32b-preview | 1st 0.37 | 1st $0.14 | $0.12 | $0.18 |
| qwen3-32b-fp8 | 1st 0.54 | 1st $0.15 | $0.10 | $0.30 |
| llama-3-3-70b-turbo-fp8 | 1st 0.52 | 1st $0.15 | $0.10 | $0.32 |
| qwen3-235b-2507 | 2nd 0.41 | 1st $0.17 | $0.07 | $0.46 |
| llama-nemotron-super-49b-v1-5 | 1st 0.44 | 1st $0.17 | $0.10 | $0.40 |
| qwen2-5-72b | 1st 0.31 | 2nd $0.19 | $0.12 | $0.39 |
| qwen3-vl-4b-fp8 | 1st 0.45 | 1st $0.23 | $0.10 | $0.60 |
| deepseek-v3-2-exp | 1st 0.62 | 1st $0.24 | $0.21 | $0.32 |
| qwen3-235b-fp8 | 1st 0.39 | 1st $0.25 | $0.13 | $0.60 |
| gpt-oss-120b-high-turbo | 1st 0.19 | 1st $0.26 | $0.15 | $0.60 |
| llama-4-maverick-fp8 | 1st 0.41 | 1st $0.26 | $0.15 | $0.60 |
| qwen3-vl-30b-a3b-fp8 | 1st 0.19 | 1st $0.26 | $0.15 | $0.60 |
| llama-3-3-70b | 0.55 | 2nd $0.27 | $0.23 | $0.40 |
| deepseek-r1-distill-qwen-32b | 1st 0.37 | 1st $0.27 | $0.27 | $0.27 |
| deepseek-v3-2 | 1st 0.61 | 1st $0.29 | $0.26 | $0.39 |
| hermes-3-llama-3-1-70b | 1st 0.30 | 1st $0.30 | $0.30 | $0.30 |
| nvidia-nemotron-nano-12b-v2-vl-fp8 | 1st 0.21 | 1st $0.30 | $0.20 | $0.60 |
| olmo-3-1-32b-instruct | 1st 0.24 | 1st $0.30 | $0.20 | $0.60 |
| llama-3-70b | 1st 0.35 | 1st $0.33 | $0.30 | $0.40 |
| qwen3-next-80b-a3b | 1st 0.30 | 2nd $0.34 | $0.09 | $1.10 |
| deepseek-v3-1-terminus-fp4 | 1st 0.42 | 1st $0.35 | $0.21 | $0.79 |
| deepseek-v3-1-fp4 | 1st 0.75 | 1st $0.35 | $0.21 | $0.79 |
| llama-3-2-90b-vision | 2nd 0.57 | 1st $0.36 | $0.35 | $0.40 |
| deepseek-v3-0324 | 0.57 | 1st $0.37 | $0.20 | $0.88 |
| llama-3-1-70b | 1st 0.42 | 1st $0.40 | $0.40 | $0.40 |
| llama-3-1-70b-turbo-fp8 | 1st 0.33 | 1st $0.40 | $0.40 | $0.40 |
| glm-4-5-air | 1st 0.17 | 2nd $0.42 | $0.20 | $1.10 |
| minimax-m2 | 1st 0.27 | 1st $0.45 | $0.25 | $1.02 |
| glm-4-6v-fp8 | 1st 0.25 | 1st $0.45 | $0.30 | $0.90 |
| qwen3-vl-235b-a22b-fp8 | 1st 0.34 | 1st $0.45 | $0.20 | $1.20 |
| deepseek-v3-dec | 1st 0.38 | 1st $0.46 | $0.32 | $0.89 |
| llama-4-maverick-turbo-fp8 | 1st 0.47 | 1st $0.50 | $0.50 | $0.50 |
| qwen3-coder-480b-turbo-fp4 | 1st 0.23 | 1st $0.51 | $0.28 | $1.20 |
| minimax-m2-1-fp8 | 1st 0.27 | 1st $0.51 | $0.28 | $1.20 |
| mixtral-8x7b | 2nd 0.32 | 1st $0.54 | $0.54 | $0.54 |
| qwen3-vl-8b-fp8 | 1st 15.00 | 1st $0.66 | $0.18 | $2.09 |
| glm-4-5 | 1st 0.36 | 1st $0.69 | $0.38 | $1.60 |
| qwen3-coder-480b-fp8 | 1st 0.28 | 2nd $0.70 | $0.40 | $1.60 |
| deepseek-r1-distill-llama-70b | 1st 0.36 | 1st $0.75 | $0.60 | $1.20 |
| glm-4-7-fp4 | 1st 0.31 | 1st $0.76 | $0.43 | $1.75 |
| glm-4-6-fp4 | 1st 0.38 | 1st $0.76 | $0.43 | $1.75 |
| qwen3-235b-a22b-2507-fp8 | 1st 0.30 | 2nd $0.77 | $0.23 | $2.39 |
| kimi-k2-0905 | 0.78 | 1st $0.80 | $0.40 | $2.00 |
| kimi-k2-thinking | 0.61 | 1st $0.85 | $0.47 | $2.00 |
| kimi-k2 | 0.76 | 1st $0.88 | $0.50 | $2.00 |
| deepseek-r1-0528 | 1st 0.38 | 1st $0.91 | $0.50 | $2.15 |
| glm-4-6-fp8 | 2nd 35.49 | 1st $0.93 | $0.60 | $1.90 |
| deepseek-r1-jan | 1st 0.35 | 1st $1.13 | $0.70 | $2.40 |
| llama-3-1-nemotron-70b | 1st 0.32 | 1st $1.20 | $1.20 | $1.20 |
| deepseek-r1-jan-turbo-fp4 | 1st 0.41 | 1st $1.50 | $1.00 | $3.00 |
llama-3-2-1b
Latency
0.44
Blended $
$0.01
Input $
$0.01
Output $
$0.01
llama-3-1-8b-turbo-fp8
Latency
0.18
Blended $
$0.02
Input $
$0.02
Output $
$0.03
llama-3-2-3b
Latency
0.46
Blended $
$0.02
Input $
$0.02
Output $
$0.02
mistral-nemo
Latency
0.19
Blended $
$0.03
Input $
$0.02
Output $
$0.04
mistral-7b
Latency
0.45
Blended $
$0.03
Input $
$0.03
Output $
$0.05
llama-3-1-8b
Latency
0.28
Blended $
$0.04
Input $
$0.03
Output $
$0.05
llama-3-8b
Latency
0.31
Blended $
$0.04
Input $
$0.03
Output $
$0.06
llama-3-2-11b-vision
Latency
1.92
Blended $
$0.05
Input $
$0.05
Output $
$0.05
gemma-3-4b
Latency
0.35
Blended $
$0.05
Input $
$0.04
Output $
$0.08
deepseek-ocr
Latency
0.17
Blended $
$0.05
Input $
$0.03
Output $
$0.10
gpt-oss-20b-high
Latency
0.18
Blended $
$0.06
Input $
$0.03
Output $
$0.14
gemma-3-12b
Latency
0.29
Blended $
$0.06
Input $
$0.04
Output $
$0.13
mistral-small-3
Latency
0.39
Blended $
$0.06
Input $
$0.05
Output $
$0.08
mistral-small-3-1
Latency
0.39
Blended $
$0.06
Input $
$0.05
Output $
$0.10
nvidia-nemotron-nano-9b-v2
Latency
0.42
Blended $
$0.07
Input $
$0.04
Output $
$0.16
devstral-small-may
Latency
0.25
Blended $
$0.07
Input $
$0.06
Output $
$0.12
gpt-oss-120b-high
Latency
0.26
Blended $
$0.08
Input $
$0.04
Output $
$0.19
qwen2-5-coder-32b
Latency
0.40
Blended $
$0.08
Input $
$0.06
Output $
$0.15
phi-4
Latency
0.31
Blended $
$0.09
Input $
$0.07
Output $
$0.14
nvidia-nemotron-3-nano
Latency
0.24
Blended $
$0.10
Input $
$0.06
Output $
$0.24
mistral-small-3-2-fp8
Latency
0.25
Blended $
$0.11
Input $
$0.07
Output $
$0.20
gemma-3-27b
Latency
0.36
Blended $
$0.11
Input $
$0.09
Output $
$0.16
qwen3-14b-fp8
Latency
0.21
Blended $
$0.12
Input $
$0.08
Output $
$0.24
devstral-small
Latency
0.26
Blended $
$0.12
Input $
$0.07
Output $
$0.28
qwen3-coder-30b-a3b-fp8
Latency
0.22
Blended $
$0.12
Input $
$0.07
Output $
$0.26
qwen3-30b-fp8
Latency
0.21
Blended $
$0.13
Input $
$0.08
Output $
$0.29
llama-4-scout
Latency
0.32
Blended $
$0.14
Input $
$0.08
Output $
$0.30
qwq-32b-preview
Latency
0.37
Blended $
$0.14
Input $
$0.12
Output $
$0.18
qwen3-32b-fp8
Latency
0.54
Blended $
$0.15
Input $
$0.10
Output $
$0.30
llama-3-3-70b-turbo-fp8
Latency
0.52
Blended $
$0.15
Input $
$0.10
Output $
$0.32
qwen3-235b-2507
Latency
0.41
Blended $
$0.17
Input $
$0.07
Output $
$0.46
llama-nemotron-super-49b-v1-5
Latency
0.44
Blended $
$0.17
Input $
$0.10
Output $
$0.40
qwen2-5-72b
Latency
0.31
Blended $
$0.19
Input $
$0.12
Output $
$0.39
qwen3-vl-4b-fp8
Latency
0.45
Blended $
$0.23
Input $
$0.10
Output $
$0.60
deepseek-v3-2-exp
Latency
0.62
Blended $
$0.24
Input $
$0.21
Output $
$0.32
qwen3-235b-fp8
Latency
0.39
Blended $
$0.25
Input $
$0.13
Output $
$0.60
gpt-oss-120b-high-turbo
Latency
0.19
Blended $
$0.26
Input $
$0.15
Output $
$0.60
llama-4-maverick-fp8
Latency
0.41
Blended $
$0.26
Input $
$0.15
Output $
$0.60
qwen3-vl-30b-a3b-fp8
Latency
0.19
Blended $
$0.26
Input $
$0.15
Output $
$0.60
llama-3-3-70b
Latency
0.55
Blended $
$0.27
Input $
$0.23
Output $
$0.40
deepseek-r1-distill-qwen-32b
Latency
0.37
Blended $
$0.27
Input $
$0.27
Output $
$0.27
deepseek-v3-2
Latency
0.61
Blended $
$0.29
Input $
$0.26
Output $
$0.39
hermes-3-llama-3-1-70b
Latency
0.30
Blended $
$0.30
Input $
$0.30
Output $
$0.30
nvidia-nemotron-nano-12b-v2-vl-fp8
Latency
0.21
Blended $
$0.30
Input $
$0.20
Output $
$0.60
olmo-3-1-32b-instruct
Latency
0.24
Blended $
$0.30
Input $
$0.20
Output $
$0.60
llama-3-70b
Latency
0.35
Blended $
$0.33
Input $
$0.30
Output $
$0.40
qwen3-next-80b-a3b
Latency
0.30
Blended $
$0.34
Input $
$0.09
Output $
$1.10
deepseek-v3-1-terminus-fp4
Latency
0.42
Blended $
$0.35
Input $
$0.21
Output $
$0.79
deepseek-v3-1-fp4
Latency
0.75
Blended $
$0.35
Input $
$0.21
Output $
$0.79
llama-3-2-90b-vision
Latency
0.57
Blended $
$0.36
Input $
$0.35
Output $
$0.40
deepseek-v3-0324
Latency
0.57
Blended $
$0.37
Input $
$0.20
Output $
$0.88
llama-3-1-70b
Latency
0.42
Blended $
$0.40
Input $
$0.40
Output $
$0.40
llama-3-1-70b-turbo-fp8
Latency
0.33
Blended $
$0.40
Input $
$0.40
Output $
$0.40
glm-4-5-air
Latency
0.17
Blended $
$0.42
Input $
$0.20
Output $
$1.10
minimax-m2
Latency
0.27
Blended $
$0.45
Input $
$0.25
Output $
$1.02
glm-4-6v-fp8
Latency
0.25
Blended $
$0.45
Input $
$0.30
Output $
$0.90
qwen3-vl-235b-a22b-fp8
Latency
0.34
Blended $
$0.45
Input $
$0.20
Output $
$1.20
deepseek-v3-dec
Latency
0.38
Blended $
$0.46
Input $
$0.32
Output $
$0.89
llama-4-maverick-turbo-fp8
Latency
0.47
Blended $
$0.50
Input $
$0.50
Output $
$0.50
qwen3-coder-480b-turbo-fp4
Latency
0.23
Blended $
$0.51
Input $
$0.28
Output $
$1.20
minimax-m2-1-fp8
Latency
0.27
Blended $
$0.51
Input $
$0.28
Output $
$1.20
mixtral-8x7b
Latency
0.32
Blended $
$0.54
Input $
$0.54
Output $
$0.54
qwen3-vl-8b-fp8
Latency
15.00
Blended $
$0.66
Input $
$0.18
Output $
$2.09
glm-4-5
Latency
0.36
Blended $
$0.69
Input $
$0.38
Output $
$1.60
qwen3-coder-480b-fp8
Latency
0.28
Blended $
$0.70
Input $
$0.40
Output $
$1.60
deepseek-r1-distill-llama-70b
Latency
0.36
Blended $
$0.75
Input $
$0.60
Output $
$1.20
glm-4-7-fp4
Latency
0.31
Blended $
$0.76
Input $
$0.43
Output $
$1.75
glm-4-6-fp4
Latency
0.38
Blended $
$0.76
Input $
$0.43
Output $
$1.75
qwen3-235b-a22b-2507-fp8
Latency
0.30
Blended $
$0.77
Input $
$0.23
Output $
$2.39
kimi-k2-0905
Latency
0.78
Blended $
$0.80
Input $
$0.40
Output $
$2.00
kimi-k2-thinking
Latency
0.61
Blended $
$0.85
Input $
$0.47
Output $
$2.00
kimi-k2
Latency
0.76
Blended $
$0.88
Input $
$0.50
Output $
$2.00
deepseek-r1-0528
Latency
0.38
Blended $
$0.91
Input $
$0.50
Output $
$2.15
glm-4-6-fp8
Latency
35.49
Blended $
$0.93
Input $
$0.60
Output $
$1.90
deepseek-r1-jan
Latency
0.35
Blended $
$1.13
Input $
$0.70
Output $
$2.40
llama-3-1-nemotron-70b
Latency
0.32
Blended $
$1.20
Input $
$1.20
Output $
$1.20
deepseek-r1-jan-turbo-fp4
Latency
0.41
Blended $
$1.50
Input $
$1.00
Output $
$3.00
FAQ
Les laboratoires d'IA (OpenAI, Anthropic, Google) développent les modèles d'IA fondamentaux. Les fournisseurs d'IA (Baseten, Together AI, Groq, DeepInfra) hébergent et mettent à disposition ces modèles via des services d'infrastructure, gérant les ressources de calcul, les API et la supervision. Les passerelles d'IA servent d'intermédiaires, offrant un accès API unifié à plusieurs fournisseurs grâce à un routage intelligent et une optimisation des coûts.
Comparez les structures tarifaires des fournisseurs, basées sur les jetons. Par exemple, notre analyse comparative a révélé qu'OpenRouter propose le tarif le plus bas (0,08 $ par million de jetons de sortie) pour Llama 4 Scout, suivi de SambaNova à 0,11 $. Nous avons également examiné les tarifs spécifiques à chaque modèle chez les fournisseurs d'IA , où des plateformes comme Baseten proposent un suivi des coûts par requête et par type de GPU pour faciliter la comparaison.
Oui, les limites de débit varient considérablement. Les fournisseurs à haut débit comme Cerebras et SambaNova sont optimisés pour les charges de travail importantes, tandis que Groq se spécialise dans les réponses à très faible latence. Des plateformes comme Baseten permettent de configurer les paramètres de mise à l'échelle automatique, mais une mauvaise configuration peut impacter les coûts et la latence. Choisissez en fonction de vos besoins : capacité de pointe ou performances stables.
Oui, plusieurs options permettent d'accéder à des modèles d'IA gratuits. De nombreux modèles open source, comme Llama, Mistral et Gemma, sont disponibles via des API gratuites sur des plateformes telles que Hugging Face, OpenRouter et Together AI, bien qu'elles soient généralement soumises à des limitations de débit et d'utilisation. Les modèles LLM open source offrent une sécurité des données renforcée, car ils peuvent être déployés sur une infrastructure privée. De plus, ils éliminent les frais de licence et les risques de dépendance vis-à-vis d'un fournisseur. Si les modèles eux-mêmes sont gratuits, les coûts de déploiement varient selon qu'ils sont exécutés localement (gratuit si votre matériel le permet), chez des fournisseurs de services gérés ou dans le cloud.
Choisissez vos fournisseurs d'IA en fonction de vos priorités. Pour un débogage détaillé, privilégiez Baseten (traçage au niveau des requêtes) ou Together AI (métriques par version). Pour une latence ultra-faible, notre test a révélé que Groq offre une latence de 0,13 s pour le premier jeton. Pour optimiser les coûts, Parasail propose une commutation GPU flexible. Pour un déploiement rapide, Fireworks AI fournit des points de terminaison API immédiats. Pour la gouvernance d'entreprise, Databricks intègre le suivi MLflow et la traçabilité des données. Envisagez l'utilisation de passerelles d'IA si vous avez besoin de flexibilité entre plusieurs fournisseurs.