NewHow we catch a seller listing the wrong model, and refund you if one does.

LLM API prices and speed, October 2026

39 numbers measured on MartLLM from 122,027 API requests to 143 models, October 19 – 31, 2026. Published November 1, 2026 and frozen: a number quoted from this page never changes. A number with too little data behind it is left out, never estimated.

API requests

122,027

Models

143

Numbers measured

39

Volume and reliability

  1. From October 19 to 31, 2026, UberLLM handled 122,027 API requests to 143 models.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-requestsMethod
  2. From October 19 to 31, 2026, 0.8% of 120,845 requests to UberLLM failed because no seller was available, a seller was rate-limited or returned a server error, or the call timed out.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-failure-rateMethod
  3. From October 19 to 31, 2026, 119,899 answered requests through UberLLM cost 71% less than the same tokens at list price.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-below-listMethod

Failure rate by model family

  1. From October 19 to 31, 2026, 0.0% of 95,369 requests to OpenAI models through UberLLM failed (no seller available, rate-limited, server error or timeout).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-failures-openaiMethod
  2. From October 19 to 31, 2026, 2.8% of 15,470 requests to DeepSeek models through UberLLM failed (no seller available, rate-limited, server error or timeout).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-failures-deepseekMethod
  3. From October 19 to 31, 2026, 0.4% of 3,444 requests to Google models through UberLLM failed (no seller available, rate-limited, server error or timeout).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-failures-googleMethod
  4. From October 19 to 31, 2026, 0.3% of 2,545 requests to Meta models through UberLLM failed (no seller available, rate-limited, server error or timeout).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-failures-meta-llamaMethod
  5. From October 19 to 31, 2026, 0.9% of 1,728 requests to MoonshotAI models through UberLLM failed (no seller available, rate-limited, server error or timeout).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-failures-moonshotaiMethod

Fastest seller per model

  1. From October 19 to 31, 2026, bedrock was the fastest of 3 sellers of Gemma 3 27B on UberLLM, with a median first token in 0.55 seconds against 0.60 seconds for openrouter (prompts under 8,000 tokens, 1,633 streamed requests).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-fastest-seller-google-gemma-3-27b-itMethod
  2. From October 19 to 31, 2026, venice was the fastest of 3 sellers of DeepSeek V4 Flash on UberLLM, with a median first token in 0.65 seconds against 0.97 seconds for bankr (prompts under 8,000 tokens, 1,081 streamed requests).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-fastest-seller-deepseek-deepseek-v4-flashMethod

Below list price

  1. From October 19 to 31, 2026, 95,232 requests to GPT-OSS 120B through UberLLM cost 96% less than the same tokens at list price ($0.18 against $4.87).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-below-list-openai-gpt-oss-120bMethod
  2. From October 19 to 31, 2026, 7,721 requests to DeepSeek V4.1 Flash 1017 through UberLLM cost 42% less than the same tokens at list price ($9.11 against $15.62).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-below-list-deepseek-deepseek-v4-1-flash-1017Method
  3. From October 19 to 31, 2026, 3,685 requests to DeepSeek V4.1 Flash through UberLLM cost 43% less than the same tokens at list price ($7.26 against $12.70).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-below-list-deepseek-deepseek-v4-1-flashMethod
  4. From October 19 to 31, 2026, 2,774 requests to Gemma 4 26B A4B IT through UberLLM cost 99% less than the same tokens at list price ($0.00061 against $0.065).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-below-list-google-gemma-4-26b-a4b-itMethod
  5. From October 19 to 31, 2026, 2,456 requests to Meta: Muse Spark 1.3 through UberLLM cost 99% less than the same tokens at list price ($2.07 against $207).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-below-list-meta-llama-muse-spark-1-3Method
  6. From October 19 to 31, 2026, 2,451 requests to DeepSeek V4 Flash through UberLLM cost 71% less than the same tokens at list price ($0.0030 against $0.010).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-below-list-deepseek-deepseek-v4-flashMethod
  7. From October 19 to 31, 2026, 1,129 requests to Kimi K3 through UberLLM cost 74% less than the same tokens at list price ($12.43 against $47.65).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-below-list-moonshotai-kimi-k3Method

Prices

  1. On October 31, 2026, GPT-OSS 120B cost $0.031 per 1M input tokens and $0.13 per 1M output tokens on UberLLM, 56% below its list price of $0.070 and $0.30.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-openai-gpt-oss-120bMethod
  2. On October 31, 2026, MiniMax M2.1 cost $0.027 per 1M input tokens and $0.11 per 1M output tokens on UberLLM, 91% below its list price of $0.30 and $1.20.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-minimax-minimax-m2-1Method
  3. On October 31, 2026, DeepSeek V4 Flash cost $0.071 per 1M input tokens and $0.14 per 1M output tokens on UberLLM, 22% below its list price of $0.091 and $0.18.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-deepseek-deepseek-v4-flashMethod
  4. On October 31, 2026, GPT 5 Mini cost $0.10 per 1M input tokens and $0.80 per 1M output tokens on UberLLM, 60% below its list price of $0.25 and $2.00.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-openai-gpt-5-miniMethod
  5. On October 31, 2026, Qwen 3 Coder 480B Turbo cost $0.31 per 1M input tokens and $1.32 per 1M output tokens on UberLLM, 12% below its list price of $0.35 and $1.50.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-qwen-qwen3-coder-turboMethod
  6. On October 31, 2026, Qwen3 Next 80B A3B cost $0.023 per 1M input tokens and $0.29 per 1M output tokens on UberLLM, 74% below its list price of $0.090 and $1.10.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-qwen-qwen3-next-80b-a3b-instructMethod
  7. On October 31, 2026, GPT-5.2 Codex cost $1.73 per 1M input tokens and $13.86 per 1M output tokens on UberLLM, 1.0% below its list price of $1.75 and $14.00.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-openai-gpt-5-2-codexMethod
  8. On October 31, 2026, Llama 3.3 70B Instruct cost $0.086 per 1M input tokens and $0.28 per 1M output tokens on UberLLM, 14% below its list price of $0.10 and $0.32.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-meta-llama-llama-3-3-70b-instructMethod
  9. On October 31, 2026, Claude Opus 4.6 cost $1.40 per 1M input tokens and $7.00 per 1M output tokens on UberLLM, 72% below its list price of $5.00 and $25.00.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-anthropic-claude-opus-4-6Method
  10. On October 31, 2026, Mistral Small 3.2 24B Instruct cost $0.052 per 1M input tokens and $0.14 per 1M output tokens on UberLLM, 30% below its list price of $0.075 and $0.20.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-price-mistralai-mistral-small-3-2-24b-instructMethod

Price gap between providers

  1. On October 31, 2026, the list price of MiniMax M2.1 input tokens ranged from $0.30 per 1M at OpenRouter to $0.35 at Venice AI across 3 providers, and UberLLM charged $0.027.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-provider-gap-minimax-minimax-m2-1Method
  2. On October 31, 2026, the list price of DeepSeek V4 Flash input tokens ranged from $0.091 per 1M at OpenRouter to $0.17 at Bankr across 4 providers, and UberLLM charged $0.071.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-provider-gap-deepseek-deepseek-v4-flashMethod
  3. On October 31, 2026, the list price of Qwen3 Next 80B A3B input tokens ranged from $0.090 per 1M at OpenRouter to $0.35 at Venice AI across 6 providers, and UberLLM charged $0.023.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-provider-gap-qwen-qwen3-next-80b-a3b-instructMethod
  4. On October 31, 2026, the list price of GPT-5.2 Codex input tokens ranged from $1.75 per 1M at Bankr to $2.19 at Venice AI across 3 providers, and UberLLM charged $1.73.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-provider-gap-openai-gpt-5-2-codexMethod
  5. On October 31, 2026, the list price of Llama 3.3 70B Instruct input tokens ranged from $0.10 per 1M at OpenRouter to $0.70 at Venice AI across 6 providers, and UberLLM charged $0.086.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-provider-gap-meta-llama-llama-3-3-70b-instructMethod
  6. On October 31, 2026, the list price of Claude Opus 4.6 input tokens ranged from $5.00 per 1M at Anthropic to $6.00 at Uncensored AI across 5 providers, and UberLLM charged $1.40.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-provider-gap-anthropic-claude-opus-4-6Method

Video price per second

  1. On October 31, 2026, Runway Gen-4.5 Text-to-Video (Venice) video cost from $0.16 per second on UberLLM (its lowest resolution).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-video-venice-runway-gen4-5-textMethod
  2. On October 31, 2026, Venice Video Gen (legacy alias) video cost from $0.16 per second on UberLLM (its lowest resolution).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-video-venice-video-genMethod
  3. On October 31, 2026, PixVerse C1 Text-to-Video (Venice) video cost from $0.070 per second on UberLLM (its lowest resolution).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-video-venice-pixverse-c1-t2vMethod
  4. On October 31, 2026, Seedance 2.0 Fast Text-to-Video (Venice) video cost from $0.059 per second on UberLLM (its lowest resolution), 61% below its list price of $0.15 per second.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-video-venice-seedance-2-fast-t2vMethod
  5. On October 31, 2026, PixVerse V5.6 Text-to-Video video cost from $0.23 per second on UberLLM (its lowest resolution).

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-video-pixverse-pixverse-v5-6-text-to-videoMethod
  6. On October 31, 2026, Kling 2.6 Pro Text-to-Video video cost from $0.055 per second on UberLLM (its lowest resolution), 64% below its list price of $0.15 per second.

    Cite this: MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10#fact-video-kling-kling-2-6-pro-text-to-videoMethod

Every number in this edition

The tables behind the facts. The same rows are in the CSV.

Prices on MartLLM

Per 1M tokens, what MartLLM charged and the official list price.

ModelInputOutputList inputList output
Aion 2.0$0.68$1.36$0.80$1.60
Aion 3.0$2.22$4.44$3.00$6.00
Aion 3.0 Mini$0.56$1.12$0.70$1.40
Aion 3.5$3.00$6.00$3.75$7.50
Aion 3.5 Mini$0.71$1.42$0.88$1.75
ByteDance Seed: Seed-2.0-Code$0.50$3.00$0.50$3.00
Claude Fable 5$4.70$23.50$10.00$50.00
Claude Fable 5.1$5.50$27.50$10.00$50.00
Claude Haiku 4.5$0.52$2.60$1.00$5.00
Claude Haiku 5.5$0.088$0.44$0.10$0.50
Claude Haiku 5.5 (batch)$0.050$0.25$0.050$0.25
Claude Opus 4.5$2.50$12.50$5.00$25.00
Claude Opus 4.6$1.40$7.00$5.00$25.00
Claude Opus 4.7$1.80$9.00$5.00$25.00
Claude Opus 4.8$2.20$11.00$5.00$25.00

List price by provider

Each provider’s official input price per 1M tokens.

ModelProviderList inputSample
Aion 2.0OpenRouter$0.802 providers
Aion 2.0Venice AI$1.002 providers
Aion 3.0OpenRouter$3.002 providers
Aion 3.0Venice AI$3.752 providers
Aion 3.0 MiniOpenRouter$0.702 providers
Aion 3.0 MiniVenice AI$0.882 providers
ByteDance Seed: Seed-2.0-CodeMorpheus$0.502 providers
ByteDance Seed: Seed-2.0-CodeOpenRouter$0.502 providers
Claude Fable 5OpenRouter$10.005 providers
Claude Fable 5Venice AI$10.005 providers
Claude Fable 5Amazon Bedrock$12.005 providers
Claude Fable 5Anthropic$12.005 providers
Claude Fable 5Bankr$12.005 providers
Claude Fable 5.1Anthropic$10.004 providers
Claude Fable 5.1Bankr$10.004 providers

Time to first token by seller

Median, streamed requests with prompts under 8,000 tokens.

ModelSellerMedianSample
Claude Sonnet 5.5openrouter5.0 s101 streamed requests
DeepSeek V4 Flashvenice0.65 s328 streamed requests
DeepSeek V4 Flashopenrouter0.76 s541 streamed requests
DeepSeek V4 Flashbankr0.97 s212 streamed requests
DeepSeek V4 Prodashscope-cn1.7 s256 streamed requests
DeepSeek V4.1 Flashdashscope-cn0.99 s2,838 streamed requests
DeepSeek V4.1 Flash 1017dashscope-cn0.94 s868 streamed requests
GLM 5.3 Flashzai1.2 s105 streamed requests
GLM-5.2dashscope-cn0.95 s160 streamed requests
GPT OSS Safeguard 20Bbedrock0.40 s453 streamed requests
GPT-OSS 120Bbedrock0.37 s378 streamed requests
GPT-OSS 20Bbedrock0.39 s110 streamed requests
Gemini 2.5 Flashgoogle0.86 s1,570 streamed requests
Gemini 3.5 Flash Liteopenrouter0.67 s195 streamed requests
Gemma 3 27Bbedrock0.55 s1,096 streamed requests

Failure rate

No seller available, rate-limited, server error or timeout.

ModelFailedSample
All models0.8%120,845 requests
OpenAI models0.0%95,369 requests
DeepSeek models2.8%15,470 requests
Google models0.4%3,444 requests
Meta models0.3%2,545 requests
MoonshotAI models0.9%1,728 requests

Below list price

What callers paid against the same tokens at list price.

ModelBelow listSample
All models71%119,899 requests
GPT-OSS 120B96%95,232 requests
DeepSeek V4.1 Flash 101742%7,721 requests
DeepSeek V4.1 Flash43%3,685 requests
Gemma 4 26B A4B IT99%2,774 requests
Meta: Muse Spark 1.399%2,456 requests
DeepSeek V4 Flash71%2,451 requests
Kimi K374%1,129 requests

Video price per second

Per second of generated video, at the model’s lowest resolution.

ModelPriceList price
Gemini Omni Flash$0.079
Gemini Omni Flash (image-to-video)$0.079$0.14
Gemini Omni Flash (reference-to-video)$0.14
Gemini Omni Flash 1.1$0.079
Grok Imagine$0.012
Grok Imagine 1.5 Image To Video Private (image-to-video)$0.090
Grok Imagine Image To Video Private (image-to-video)$0.060
Grok Imagine Reference To Video Private (reference-to-video)$0.060
Grok Imagine Text-to-Video$0.012$0.088
Grok Imagine Video 1.5$0.012
Hailuo 02 Pro$0.048
Hailuo 02 Standard Image to Video$0.010$0.017
Hailuo 02 Standard Text to Video$0.025$0.045
Hailuo 2.3 Pro$0.045
Hailuo 2.3 Standard$0.025

How we measure

Every number comes from MartLLM’s own records for October 19 – 31, 2026: the 122,027 API requests callers sent to MartLLM, the log of which seller answered each one, and a daily copy of the public model catalog. Nothing is estimated, sampled from another source or edited by hand.

Prices are as listed on October 31, 2026, the last day of the month we priced: what MartLLM charged per 1M tokens (or per second of video) and each model’s official list price. A price change is reported when the first and last price of the month are at least 14 days apart and differ by 5% or more. A video price is per second of video at the model’s lowest resolution; a job costs that rate times its length.

Time to first token is the median (and 90th percentile) over streamed requests with prompts under 8,000 tokens, because a long prompt takes longer to read on every seller. A model needs 300 such requests and a seller 100.

A request failed when no seller answered it: none was available, the seller was rate-limited or returned a server error, or the call timed out. Requests refused because of the caller (a bad request, a missing key, no credit) and calls the caller cancelled are not counted either way. A family needs 1,000 requests.

Below list price compares what callers paid MartLLM for 1,000 or more answered requests to a model with what the same tokens cost at the model’s official list price on the day the edition was built.

An edition is published on the 1st of the next month, only when the month had at least 7 days of data and 20 numbers passed their minimum. It is then frozen.

Cite this edition

MartLLM, "LLM API price and speed stats, October 2026", martllm.com/stats/2026-10

Free to reuse with a link (CC BY 4.0). Data: CSV · JSON