AI Calculator Pro
Monthly recap

What changed in LLM pricing — September 2026

In September 2026, we recorded 53 price changes (26 cuts, 27 increases) and 34 new models across the LLM providers we track. The biggest cut was Mistral Small 3.1 24B Instruct input, down 72% ($0.351 → $0.100 per 1M tokens).

The median cut this month was 17%. Figures are from our daily re-check of official provider pricing; see the live tracker for the full history.

53
Price changes
26
Price drops
27
Price increases
34
New models

Price cuts in September 2026

Models that got cheaper, biggest cut first. Click a column to sort.

Mistral Small 3.1 24B InstructInput$0.351/1M$0.100/1M-72%15 September 2026Mistral AI
Qwen3.8 2.4T A95BCached input$0.500/1M$0.200/1M-60%17 September 2026Alibaba
gemma-4-31b-itCached input$0.020/1M$0.010/1M-50%11 September 2026Google
Mistral Small 3.1 24B InstructOutput$0.555/1M$0.300/1M-46%15 September 2026Mistral AI
deepseek-v4-1-flashOutput$0.600/1M$0.400/1M-33%15 September 2026DeepSeek
deepseek-v4-1-flashInput$0.150/1M$0.100/1M-33%15 September 2026DeepSeek
Qwen3 Coder Next (Bedrock)Output$1.80/1M$1.20/1M-33%12 September 2026AWS Bedrock
Llama-3.3-70B-InstructInput$0.135/1M$0.100/1M-26%3 September 2026Meta

Price increases in September 2026

mistral-medium-3-5Cached input$0.139/1M$0.750/1M+440%15 September 2026Mistral AI
Qwen3.8 27BCached input$0.010/1M$0.050/1M+400%17 September 2026Alibaba
Google Gemma 3 12B (Bedrock)Output$0.100/1M$0.290/1M+190%11 September 2026AWS Bedrock
Qwen3 Coder Next (Bedrock)Input$0.220/1M$0.500/1M+127%12 September 2026AWS Bedrock
DeepSeek V4 FlashOutput$0.280/1M$0.600/1M+114%11 September 2026DeepSeek
Qwen3.8 2.4T A95BCached input$0.250/1M$0.500/1M+100%15 September 2026Alibaba
Qwen3.7 FlashCached input$0.003/1M$0.006/1M+100%1 September 2026Alibaba
Google Gemma 3 27B Instruct (Bedrock)Input$0.120/1M$0.230/1M+92%11 September 2026AWS Bedrock

Models launched in September 2026

Cite this recap

Monthly recaps are generated from our open price-change dataset (CC BY 4.0). See the recap JSON, the full change dataset, or the RSS feed.