OpenRouter provider price spread

Spread by model endpoints read in Switzerland on 9 Nov 2026, widest spread first

The other 89 models have a spread of 1.8× or less: each model page lists its endpoints.

Observed unpinned cost what OpenRouter charged for requests without a pinned provider

1359 short requests sent without a pinned provider on 25 Oct 2026 – 9 Nov 2026 (16 dates), in each routing mode, for 11 models. Charged / listing is the sum of the OpenRouter generation receipts divided by the price of the same tokens at the listing price of that date. The model page shows the amounts and the endpoint prices.

Routing modes as documented by OpenRouter (OpenRouter provider routing, read 25 Oct 2026): the default request has no provider field; :floor and :nitro are suffixes of the model name. Each request asks for one short line with an answer cap of 32 tokens. The counts are for these requests on these dates only.

By provider lowest- and highest-priced endpoint, readings of 3 Nov 2026 – 9 Nov 2026 (7 dates)

For every model with endpoints from two or more providers at different prices, each nightly reading names the provider of the lowest-priced and of the highest-priced endpoint for the workload above. The table counts these per provider over the readings of the last 7 days: 7 readings. A provider with several endpoints of one model counts once per reading; equal prices count for each provider.

15 providers more with at least one count are not shown. 13 other providers had endpoints in these readings and were neither the lowest- nor the highest-priced for the workload.

Price moves 20% or more between two readings

28 models with a price move of 20% or more between the readings of 8 Nov 2026 and 9 Nov 2026:

• deepseek/deepseek-chat
  • Listing price $0.2574 / $1.0287 → $0.32 / $0.89 (input +24%, output −13%).
• deepseek/deepseek-v4-flash
  • Relace (relace/fp4): $0.0413 / $0.45 → $0.05 / $0.45 (input +21%, output 0%).
  • StreamLake (streamlake/fp8): $0.04354 / $0.08708 → $0.0826 / $0.1652 (input +90%, output +90%).
  • Wafer (wafer/mxfp4): $0.60 / $0.17 → $0.12 / $0.17 (input −80%, output 0%).
  • Baidu (baidu/fp8): $0.04424 / $0.08848 → $0.14 / $0.28 (input +216%, output +216%).
  • Lowest-priced endpoint for the workload: Relace (relace/fp4) at $0.59 per 1,000 tasks, before StreamLake (streamlake/fp8) at $0.36.
• deepseek/deepseek-v4-flash-0731
  • Listing price $0.059 / $0.30 → $0.14 / $0.30 (input +137%, output 0%).
  • StreamLake (streamlake/fp8): $0.088 / $0.264 → $0.132 / $0.396 (input +50%, output +50%).
  • Baidu (baidu/fp8): $0.04444 / $0.13332 → $0.44 / $1.32 (input +890%, output +890%).
  • PrimeIntellect (primeintellect/fp4): $0.09 / $0.23 → $0.44 / $1.32 (input +389%, output +474%).
  • Lowest-priced endpoint for the workload: Wafer (wafer/us) at $0.47 per 1,000 tasks, before Baidu (baidu/fp8) at $0.39.
• deepseek/deepseek-v4-pro
  • StreamLake (streamlake/fp8): $0.454314 / $0.908628 → $0.95526 / $1.91052 (input +110%, output +110%).
  • Baidu (baidu/fp8): $0.45292 / $0.90584 → $1.69 / $3.38 (input +273%, output +273%).
  • Lowest-priced endpoint for the workload: Relace (relace/fp4) at $7.32 per 1,000 tasks, before Baidu (baidu/fp8) at $3.76.
• deepseek/deepseek-v4-pro-0813
  • Listing price $0.103 / $4.11 → $1.32 / $3.96 (input +1182%, output −4%).
  • Baidu (baidu/fp8): $0.65868 / $1.97604 → $1.32 / $3.96 (input +100%, output +100%).
  • DeepSeek (deepseek): $0.66 / $1.98 → $1.32 / $3.96 (input +100%, output +100%).
• deepseek/deepseek-v4.1-flash
  • Listing price $0.045 / $0.50 → $0.21 / $0.84 (input +367%, output +68%).
  • PrimeIntellect (primeintellect): $0.156 / $1.20 → $0.039 / $1.20 (input −75%, output 0%).
  • Baidu (baidu/fp8): $0.0507 / $0.2028 → $0.30 / $1.20 (input +492%, output +492%).
• deepseek/deepseek-v4.1-flash-1017
  • Listing price $0.30 / $1.20 → $0.165 / $0.68 (input −45%, output −43%).
  • PrimeIntellect (primeintellect): $0.1092 / $1.20 → $0.08736 / $1.20 (input −20%, output 0%).
  • Baidu (baidu/fp8): $0.09772 / $0.39437 → $0.28 / $1.13 (input +187%, output +187%).
  • DeepSeek (deepseek): $0.15 / $0.60 → $0.30 / $1.20 (input +100%, output +100%).
• deepseek/deepseek-v4.1-flash:batch
  • Listing price $0.112 / $0.336 → $0.011 / $0.20 (input −90%, output −40%).
• google/gemma-4-26b-a4b-it
  • Lowest-priced endpoint for the workload: PrimeIntellect (primeintellect/bf16) at $0.32 per 1,000 tasks, before Darkbloom (darkbloom) at $0.37.
• inclusionai/ling-3.1-flash
  • Listing price $0.075 / $0.225 → $0.045 / $0.135 (input −40%, output −40%).
• minimax/minimax-m2.5
  • Listing price $0.27 / $0.95 → $0.30 / $1.20 (input +11%, output +26%).
• moonshotai/kimi-k2.5
  • Listing price $0.50 / $2.50 → $0.532 / $3.325 (input +6%, output +33%).
• moonshotai/kimi-k2.6
  • Baidu (baidu/fp4): $0.3895 / $1.64 → $0.95 / $4.00 (input +144%, output +144%).
  • Lowest-priced endpoint for the workload: Chutes (chutes/int4) at $3.75 per 1,000 tasks, before Baidu (baidu/fp4) at $3.66.
• moonshotai/kimi-k3
  • Listing price $0.192 / $16.00 → $2.2335 / $15.00 (input +1063%, output −6%).
  • Io Net (io-net/fp4): $2.30 / $14.00 → $1.50 / $14.00 (input −35%, output 0%).
• moonshotai/kimi-k3:batch
  • Listing price $2.28 / $11.40 → $1.084 / $6.00 (input −52%, output −47%).
• moonshotai/kimi-k3.1-pro
  • Relace (relace/fp4): $0.377 / $7.9515 → $0.19 / $8.9276 (input −50%, output +12%).
  • Wafer (wafer/us): $0.399 / $10.00 → $0.20 / $10.00 (input −50%, output 0%).
• nvidia/nemotron-3-super-120b-a12b
  • Listing price $0.30 / $2.40 → $0.075 / $0.40 (input −75%, output −83%).
• nvidia/nemotron-3.5-super-vl
  • Io Net (io-net/nvfp4): $0.15 / $0.50 → $0.09 / $0.30 (input −40%, output −40%).
• openai/gpt-oss-120b
  • Listing price $0.026 / $0.096 → $0.10 / $0.50 (input +285%, output +421%).
  • PrimeIntellect (primeintellect/fp4): $0.027 / $0.10 → $0.039 / $0.083 (input +44%, output −17%).
  • Lowest-priced endpoint for the workload: Near AI (near-ai/fp4) at $0.29 per 1,000 tasks, before PrimeIntellect (primeintellect/fp4) at $0.25.
• openai/gpt-oss-20b
  • PrimeIntellect (primeintellect/fp4): $0.004 / $0.017 → $0.02 / $0.03 (input +400%, output +76%).
  • Lowest-priced endpoint for the workload, PrimeIntellect (primeintellect/fp4): $0.038 → $0.16 per 1,000 tasks (+327%).
• qwen/qwen3.5-9b
  • PrimeIntellect (primeintellect/bf16): $0.029 / $0.043 → $0.061 / $0.09 (input +110%, output +109%).
  • Lowest-priced endpoint for the workload, PrimeIntellect (primeintellect/bf16): $0.23 → $0.49 per 1,000 tasks (+110%).
• qwen/qwen3.6-27b
  • Listing price $0.45 / $2.70 → $0.32 / $3.20 (input −29%, output +19%).
• qwen/qwen3.8-27b
  • Mancer 2 (mancer/fp8): $0.175 / $2.50 → $0.15 / $1.25 (input −14%, output −50%).
  • PrimeIntellect (primeintellect/fp8): $0.021 / $1.159 → $0.45 / $3.20 (input +2043%, output +176%).
  • Lowest-priced endpoint for the workload: Chutes (chutes/fp8) at $0.91 per 1,000 tasks, before PrimeIntellect (primeintellect/fp8) at $0.73.
• z-ai/glm-5.1
  • Baidu (baidu/fp8): $0.8988 / $2.8248 → $1.40 / $4.40 (input +56%, output +56%).
  • Lowest-priced endpoint for the workload: Chutes (chutes/fp8) at $8.02 per 1,000 tasks, before Baidu (baidu/fp8) at $7.97.
• z-ai/glm-5.2
  • Listing price $0.006 / $6.80 → $1.12 / $4.40 (input +18567%, output −35%).
  • StreamLake (streamlake/fp8): $0.28 / $0.88 → $0.61306 / $1.92676 (input +119%, output +119%).
  • Morph (morph/fp8): $0.74 / $3.00 → $1.245 / $2.542 (input +68%, output −15%).
  • Baidu (baidu/fp8): $0.189 / $0.594 → $1.40 / $4.40 (input +641%, output +641%).
  • Lowest-priced endpoint for the workload: Wafer (wafer/us) at $3.44 per 1,000 tasks, before Baidu (baidu/fp8) at $1.68.
• z-ai/glm-5.3
  • Listing price $0.0699 / $1.40 → $1.00 / $4.20 (input +1331%, output +200%).
  • Morph (morph/fp8): $0.74 / $3.00 → $1.245 / $2.542 (input +68%, output −15%).
• z-ai/glm-5.3:batch
  • Listing price $0.45 / $2.00 → $0.37 / $1.50 (input −18%, output −25%).
  • Morph (morph/fp8): $0.37 / $1.50 → $0.623 / $1.271 (input +68%, output −15%).
  • Lowest-priced endpoint for the workload: DeepInfra (deepinfra/fp4) at $4.29 per 1,000 tasks, before Morph (morph/fp8) at $3.45.
• z-ai/glm-5.3-flash
  • OpenInference (open-inference/fp4): $0.06 / $0.24 → $0.06 / $0.30 (input 0%, output +25%).
  • PrimeIntellect (primeintellect): $0.15 / $0.50 → $0.105 / $0.50 (input −30%, output 0%).

Moves count changes of the listing price, of the same endpoint’s price, and of the lowest-priced endpoint for the workload. Endpoints that appear or leave are not price moves. Feed: price moves (Atom).

How this is read

The source is the OpenRouter endpoints list of every model on OpenRouter that the model picker compares: provider, endpoint tag, precision, context and price per 1M input and output tokens, with the date it was read. The listing price is the one OpenRouter shows for the model. Costs are the published prices multiplied by the workload; no endpoint weights are applied. Observed unpinned cost comes from the generation receipts of our own requests and appears once a model has them on two dates or more.