Home Models Compare Local Models Pricing Scorecards Evals OpenClaw Methodology

Pricing guide

Cheapest AI Models That Still Hold Up

Cheap is only useful if the model is still competent. These are the low-cost options that don’t completely suck.

ModelInputOutputTotal price signalBest for
GPT-OSS-120B
OpenAI
$0$0$0.00Self-hosted
Grok 4.1 Fast
xAI
$0.2$0.5$0.70Long context
DeepSeek V3
DeepSeek
$0.27$1.1$1.37Budget coding
MiniMax M2.5
MiniMax
$0.3$1.2$1.50Real-world productivity
GLM-5
Zhipu AI
$0.5$2$2.50Bilingual (CN/EN)
Kimi K2.5
Moonshot AI
$0.6$2.5$3.10Visual coding
Gemini 3.1 Pro
Google
$1.25$5$6.25Multimodal tasks
Qwen 3 Max
Alibaba
$1.2$6$7.20Multilingual
Mistral Large 3
Mistral
$2$6$8.00European compliance
Llama 4 (405B)
Meta
$2$8$10.00Self-hosted
GPT-5.2
OpenAI
$1.75$14$15.75General-purpose
GPT-5.4
OpenAI
$2.5$15$17.50Coding
Claude Sonnet 4.6
Anthropic
$3$15$18.00Balanced performance
GPT-5.3 Codex
OpenAI
$3$15$18.00Coding-focused tasks
Claude Opus 4.6
Anthropic
$5$25$30.00Complex reasoning

Best ultra-budget option

Grok 4.1 Fast is stupidly cheap for a 2M-context model and works well for long-context + search-heavy workloads.

Best value frontier model

MiniMax M2.5 is the best balance of serious quality and low token cost.

Best bilingual value

GLM-5 is still one of the strongest value models if Chinese/English matters.

When cheap is the right choice