AI API prices in September 2026: what changed

In September 2026 (full month), 49 headline prices moved between the start and the end of the month, across 28 offers from 2 sellers: 31 cuts and 18 increases. 485 new offers were first read from sellers we already followed, 17 of them sold by the model's own vendor. 72 offers stopped being listed. 5 models received a dated end of support or retirement.

Largest price cuts

6 more are listed on every recorded change.

Largest price increases

New offers

Offers first read this month from sellers Bertrande already followed, with their current output price. A seller we start following is not counted, so a new collector does not pass for a wave of launches. Sold by the model's own vendor:

Resellers added offers too: Novita AI (162), DeepInfra (116), OpenRouter (74), Cloudflare (40), Nebius Token Factory (25), Fireworks AI (19), Baseten (17), Groq (7), Replicate (4), Cerebras (2), Mistral AI (1), OVHcloud (1).

Retired, ended or no longer listed

• Sonar: vendor support ended on 2026-09-27, as the vendor states on its own page.
• Sonar Deep Research: vendor support ended on 2026-09-27, as the vendor states on its own page.
• Sonar Pro: vendor support ended on 2026-09-27, as the vendor states on its own page.
• Sonar Reasoning Pro: vendor support ended on 2026-09-27, as the vendor states on its own page.
• Claude Sonnet 4.5: retired by the vendor on 2026-09-29, as the vendor states on its own page.

72 offers stopped being listed by their seller this month. Their last prices stay published, dated, on their pages.

• DeepSeek: DeepSeek V4 Flash 0731 (batch) (sold by OpenRouter)
• DeepSeek: DeepSeek V4 Flash Vision Exp (batch) (sold by OpenRouter)
• DeepSeek: DeepSeek V4 Pro 0813 (batch) (sold by OpenRouter)
• GLM 5.2 (hosted by Mistral)
• Google: Gemma 4 31B (batch) (sold by OpenRouter)
• Meta: Muse Glimmer 30B (batch) (sold by OpenRouter)
• MiniMax: MiniMax M3 (batch) (sold by OpenRouter)
• MiniMaxAI/MiniMax-M2.7-Turbo (on DeepInfra)
• NVIDIA: Nemotron 3 Ultra (batch) (sold by OpenRouter)
• OpenAI GPT Latest (sold by OpenRouter)
• OpenAI: GPT-4 Turbo Preview (sold by OpenRouter)
• Qwen/Qwen3.6-27B (on Groq)
• Qwen3.7 Plus (on Fireworks AI)
• Qwen: Qwen3.5-9B (batch) (sold by OpenRouter)
• Qwen: Qwen3.8 2.4T A95B (batch) (sold by OpenRouter)
• Thinking Machines: Inkling (batch) (sold by OpenRouter)
• Thinking Machines: Inkling Small (batch) (sold by OpenRouter)
• XiaomiMiMo/MiMo-V2.6-Flash-RL (on DeepInfra)
• Z.ai: GLM 5.2 (batch) (sold by OpenRouter)
• baidu/ernie-4.5-21B-a3b-thinking (on Novita AI)
• baidu/ernie-4.5-300b-a47b-paddle (on Novita AI)
• baidu/ernie-4.5-vl-28b-a3b (on Novita AI)
• baidu/ernie-4.5-vl-28b-a3b-thinking (on Novita AI)
• deepseek-v4-flash
• deepseek-v4-flash-vision-exp
• deepseek/deepseek-ocr (on Novita AI)
• deepseek/deepseek-prover-v2-671b (on Novita AI)
• deepseek/deepseek-r1-distill-qwen-14b (on Novita AI)
• deepseek/deepseek-r1-distill-qwen-32b (on Novita AI)
• deepseek/deepseek-v3-0324 (on Novita AI)
• deepseek/deepseek-v3-turbo (on Novita AI)
• deepseek/deepseek-v3/community (on Novita AI)
• deepseek/deepseek-v4.1-flash-z (on Novita AI)
• elephant (on Novita AI)
• elevenlabs/music (on Replicate)
• google/gemma-4-31B (on Cerebras)
• google/gemma-4-E4B-it (on DeepInfra)
• hermes/mercury-oss-120b (on Novita AI)
• hermes/mercury-oss-20b (on Novita AI)
• humbleworth/price-predict-v1 (on Replicate)

How this page is built: every figure comes from the registry and links to the offer page, where each value carries the day it was read, its source and a fingerprint. A price counts as moved only if it differs between the start and the end of the month and at least one real movement was confirmed in between; prices that bounce back are left out, and so are aliases such as "latest" models, whose price changes when the alias moves to another model. See the methodology.