AI Model Pricing & Benchmarks

Prix réels sur OpenRouter + Benchmarks agentic 2026

Prix par 1M tokens | SWE-bench, MCP-Atlas, Terminal-Bench
Moins cher Plus cher

OpenAI GPT-5.5

2 providers

Moins cher / 1M

Input$5.00
Output$30.00
OpenAI / OpenAI

Plus cher / 1M

Input$5.00
Output$30.00
OpenAI / OpenAI

Combiné in+out / 1M

Min$35.00
Max$35.00
×1.0 in / ×1.0 out
Agentic
Tool calling, structured outputs, 1M ctx, reasoning
Spread: in 1.0x out 1.0xCombiné/1M: $35.00$35.00
ProviderInputOutputCombiné
OpenAI$5.00$30.00$35.00
Azure$5.00$30.00$35.00

Claude Opus 4.7

3 providers

Moins cher / 1M

Input$5.00
Output$25.00
Google / Google

Plus cher / 1M

Input$5.00
Output$25.00
Google / Google

Combiné in+out / 1M

Min$30.00
Max$30.00
×1.0 in / ×1.0 out
Agentic
Long-running async agents, frontier coding, 1M ctx, computer use
Spread: in 1.0x out 1.0xCombiné/1M: $30.00$30.00
ProviderInputOutputCombiné
Google$5.00$25.00$30.00
Bedrock$5.00$25.00$30.00
Anthropic$5.00$25.00$30.00

Claude Sonnet 4.6

4 providers

Moins cher / 1M

Input$3.00
Output$15.00
Anthropic / Anthropic

Plus cher / 1M

Input$3.00
Output$15.00
Anthropic / Anthropic

Combiné in+out / 1M

Min$18.00
Max$18.00
×1.0 in / ×1.0 out
Agentic
Frontier coding + agents, 1M ctx, best cost-perf Sonnet
Spread: in 1.0x out 1.0xCombiné/1M: $18.00$18.00
ProviderInputOutputCombiné
Anthropic$3.00$15.00$18.00
Bedrock$3.00$15.00$18.00
Google$3.00$15.00$18.00
Azure$3.00$15.00$18.00

DeepSeek V4 Pro

5 providers

Moins cher / 1M

Input$0.4350
Output$0.8700
DeepSeek / DeepSeek

Plus cher / 1M

Input$2.10
Output$4.40
Together / Together

Combiné in+out / 1M

Min$1.30
Max$6.50
×4.8 in / ×5.1 out
Agentic
1.6T MoE, 49B active, 1M ctx, reasoning + coding
Spread: in 4.8x out 5.1xCombiné/1M: $1.30$6.50
ProviderInputOutputCombiné
DeepSeek$0.435$0.870$1.30
GMICloud$1.39$2.78$4.18
AtlasCloud$1.70$3.40$5.10
SiliconFlow$1.74$3.48$5.22
Together$2.10$4.40$6.50

Qwen 3.6 Plus

1 provider

Moins cher / 1M

Input$0.3250
Output$1.95
Alibaba / Alibaba

Plus cher / 1M

Input$0.3250
Output$1.95
Alibaba / Alibaba

Combiné in+out / 1M

Min$2.27
Max$2.27
×1.0 in / ×1.0 out
Agentic
Hybrid attn + MoE, 1M ctx, reasoning + tool use
Spread: in 1.0x out 1.0xCombiné/1M: $2.27$2.27
ProviderInputOutputCombiné
Alibaba$0.325$1.95$2.27

Kimi K2.6

7 providers

Moins cher / 1M

Input$0.7400
Output$3.49
Io Net / Io Net

Plus cher / 1M

Input$1.20
Output$4.50
Together / Together

Combiné in+out / 1M

Min$4.23
Max$5.70
×1.6 in / ×1.3 out
Agentic
Agent Swarm (100 sub-agents, 1500 parallel tools), long-horizon
Spread: in 1.6x out 1.3xCombiné/1M: $4.23$5.70
ProviderInputOutputCombiné
Io Net$0.740$3.49$4.23
DeepInfra$0.750$3.50$4.25
Parasail$0.800$3.50$4.30
Moonshot$0.950$4.00$4.95
Cloudflare$0.950$4.00$4.95
Together$1.20$4.50$5.70

MiniMax M2.7

4 providers

Moins cher / 1M

Input$0.3000
Output$1.20
Fireworks / Fireworks

Plus cher / 1M

Input$0.6000
Output$2.40
SambaNova / SambaNova

Combiné in+out / 1M

Min$1.50
Max$3.00
×2.0 in / ×2.0 out
Agentic
Autonomous productivity, 192K ctx, coding-first
Spread: in 2.0x out 2.0xCombiné/1M: $1.50$3.00
ProviderInputOutputCombiné
Fireworks$0.300$1.20$1.50
Together$0.300$1.20$1.50
Minimax$0.300$1.20$1.50
SambaNova$0.600$2.40$3.00

GLM-5.1

6 providers

Moins cher / 1M

Input$1.05
Output$3.50
Chutes / Chutes

Plus cher / 1M

Input$1.75
Output$5.50
Venice / Venice

Combiné in+out / 1M

Min$4.55
Max$7.25
×1.7 in / ×1.6 out
Agentic
SWE-Pro 58.4%, long-horizon tasks, 202K ctx
Spread: in 1.7x out 1.6xCombiné/1M: $4.55$7.25
ProviderInputOutputCombiné
Chutes$1.05$3.50$4.55
DeepInfra$1.05$3.50$4.55
Z.AI$1.40$4.40$5.80
Together$1.40$4.40$5.80
SiliconFlow$1.40$4.40$5.80
Venice$1.75$5.50$7.25

Grok 4.3

1 provider

Moins cher / 1M

Input$1.25
Output$2.50
xAI / xAI

Plus cher / 1M

Input$1.25
Output$2.50
xAI / xAI

Combiné in+out / 1M

Min$3.75
Max$3.75
×1.0 in / ×1.0 out
Agentic
Reasoning model, agentic workflows, 1M ctx, image input
Spread: in 1.0x out 1.0xCombiné/1M: $3.75$3.75
ProviderInputOutputCombiné
xAI$1.25$2.50$3.75
★ TopBonMoyenN/A
Modèle SWE-VerSWE ProMCPTermFinCoding Notes
OpenAI GPT-5.5
82.6%
58.6%
82.7%
1520
Terminal-Bench 82.7%, Expert-SWE 73.1%
Claude Opus 4.7
87.6%
64.3%
77.3%
64.4%
1548
#1 SWE-Pro 64.3%, MCP-Atlas 77.3%, GPQA 94.2%
Claude Sonnet 4.6
79.6%
53.4%
75.8%
60.1%
1530
79.6% SWE-Ver at ~50% Opus price. Arena 1530
DeepSeek V4 Pro
~66%
~66% SWE-Ver at 10x cheaper than frontier
Qwen 3.6 Plus
No public agentic scores yet
Kimi K2.6
~70%
~70% SWE-Ver. 100 sub-agents, 1500 parallel calls
MiniMax M2.7
~72.5
~72.5% SWE-Ver. Best overall value
GLM-5.1
58.4%
SWE-Pro 58.4% best among non-OpenAI/Anthropic
Grok 4.3
No public agentic benchmark scores yet

Recommandations pour OpenClaw

Codage pur — meilleur résultat

#1 SWE-bench Pro (64.3%), meilleur tool invocation (77.3%), +14% multi-step vs Opus 4.6. Choix premium pour Codex, refactoring, features.

Combined 1M: $30.00
Best perf

Codage pur — meilleur rapport

79.6% SWE-Verified à ~50% prix d'Opus. Idéal pour le quotidien: review PR, fixes, petites features.

Combined 1M: $18.00
Best value

Budget — prix agentique

~72.5% SWE-Ver prix imbattable. Coding autonome + amélioration continue.

Combined 1M: $1.50
Cheap agentic72.5% SWE

Budget ultra-serré

~66% SWE-Ver prix plancher. 1.6T MoE, 49B params. Exploration, questions.

Combined 1M: $1.305
Prix plancher

CLI + Shell

Terminal-Bench 82.7% (meilleur), Expert-SWE 73.1%. Commands shell, debug, fichiers.

Combined 1M: $35.00
Terminal #1

Coût par session OpenClaw (50K in + 50K out = 100K tokens)

Modèle/100KCombined /1MUsage recommandé
OpenAI GPT-5.5$ 3.5000$35.00Ultra-premium
Claude Opus 4.7$ 3.0000$30.00Premium
Claude Sonnet 4.6$ 1.8000$18.00Premium
DeepSeek V4 Pro$ 0.1305$1.30Budget
Qwen 3.6 Plus$ 0.2275$2.27Budget
Kimi K2.6$ 0.4230$4.23Quotidien
MiniMax M2.7$ 0.1500$1.50Budget
GLM-5.1$ 0.4550$4.55Quotidien
Grok 4.3$ 0.3750$3.75Quotidien

Stratégie recommandée :
Default : Sonnet 4.6 — 90% des tâches
Complexe : Opus 4.7 — quand Sonnet bloque
Fallback : GPT-5.5 — shell, debug
Budget : MiniMax/DeepSeek — tâches répétitives