InfratoApp / LLM prices

What a million tokens costs

509 models priced across 16 sellers — 11 who resell, and 5 makers publishing their own list price (Anthropic, DeepSeek, Mistral AI, Moonshot AI, OpenAI). Where a maker sells the model itself and we have not read its rate, the row carries its price page and a ○. The lowest input rate on record is $0.0170 per million, on ibm-granite/granite-4.0-h-micro.

23 free to call →356 with cached rates →How we read these prices →

Same model, different price · input per 1M

Sold by several sellers

509

models priced

356

publish a cached rate

16

sellers

124

sold by 2+ sellers

2026-10-27

most recent read

By seller

Every seller with a price on record, each with its own page.

Same model, different seller

The same model varies up to 41.1× between sellers — z-ai/glm-5.2 is $0.0341 per million input tokens at OpenRouter and $1.4 at FriendliAI.

z-ai/glm-5.241.1× · $0.0341 OpenRouter → $1.4 FriendliAI
z-ai/glm-5.320.0× · $0.0699 OpenRouter → $1.4 FriendliAI
moonshotai/kimi-k315.9× · $0.189 OpenRouter → $3 Fireworks AI
deepseek/deepseek-v3.210.7× · $0.28 OpenRouter → $3 SambaNova
openai/gpt-oss-120b7.0× · $0.05 Novita AI → $0.35 Cerebras Inference
inclusionai/ling-3.1-flash4.0× · $0.075 OpenRouter → $0.3 Novita AI

More models than this appear in two catalogues; only these carry an input and an output rate from each seller, which is what a comparison needs.

Every model we hold both halves for

ModalitiesSellersRead from

ibm-granite/granite-4.0-h-microIBM

131K$0.017$0.112—OpenRouter only2026-10-26

openai/gpt-oss-20bOpenAI

131K$0.018$0.090$0.0092026-10-27

~deepseek/deepseek-flash-latestDeepseek

1M$0.020$1.00$0.0152026-10-27

meta-llama/llama-3.1-8b-instructMeta

131K$0.020$0.050—2026-10-22

inclusionai/ling-3.0-flashInclusionAI

262K$0.021$0.063$0.0042026-10-27

openai/gpt-oss-20b:batchOpenAI

131K$0.024$0.112—2026-10-27

Nnex-agi/nex-n2.5-miniNex-agi

262K$0.025$0.100$0.003OpenRouter only2026-10-27

openai/gpt-5-nano:batchOpenAI

400K$0.025$0.200$0.0032026-10-27

meta-llama/llama-3.2-1b-instructMeta

60K$0.027$0.201—OpenRouter only2026-10-27

~z-ai/glm-flash-latestZ.ai

1M$0.029$0.080$0.0292026-10-27

mistralai/mistral-nemoMistral

131K$0.029$0.030—2026-10-27

openai/gpt-oss-120b:batchOpenAI

131K$0.030$0.136—2026-10-27

deepseek/deepseek-ocr-2Deepseek

○—$0.030$0.030—2026-10-22

Iinference-net/schematron-v2-turboInference-net

128K$0.030$0.150$0.030OpenRouter only2026-10-27

qwen/qwen3.7-flashAlibaba

1M$0.030$0.130$0.006OpenRouter only2026-10-27

amazon/nova-micro-v1Amazon

128K$0.035$0.140—2026-10-27

z-ai/autoglm-phone-9b-multilingualZ.ai

○—$0.035$0.138—2026-10-22

cohere/command-r7b-12-2024Cohere

128K$0.037$0.150—2026-10-27

Iinception/mercury-2.5Inception

260K$0.040$0.150$0.004OpenRouter only2026-10-27

Ssao10k/l3-lunaris-8bSao10k

8K$0.040$0.050—OpenRouter only2026-10-27

Sscaledown/extractScaledown

1M$0.040free—OpenRouter only2026-10-27

Sscaledown/summarizeScaledown

1M$0.040free—OpenRouter only2026-10-27

tencent/hy-mt2-1.8bTencent

8K$0.044$0.177—OpenRouter only2026-10-27

~deepseek/deepseek-v4-flash-latestDeepseek

1M$0.048$0.157$0.0282026-10-27

anthropic/claude-haiku-5.5:batchAnthropic

1M$0.050$0.250$0.0052026-10-27

deepseek/deepseek-v4-flashDeepseek

1M$0.050$0.450$0.0502026-10-27

deepseek/deepseek-v4.1-flash:batchDeepseek

1M$0.050$0.150$0.0042026-10-27

google/gemini-2.5-flash-lite:batchGoogle

1M$0.050$0.200$0.0102026-10-21

google/gemma-3-12b-itGoogle

131K$0.050$0.150—2026-10-27

google/gemma-3-4b-itGoogle

131K$0.050$0.100$0.0202026-10-27

Iinference-net/schematron-v2-smallInference-net

128K$0.050$0.230$0.050OpenRouter only2026-10-27

Lliquid/lfm-2.5-24b-a2bLiquid

128K$0.050$0.700$0.025OpenRouter only2026-10-27

meta-llama/llama-3.2-3b-instructMeta

131K$0.050$0.330—OpenRouter only2026-10-27

mistralai/mistral-small-24b-instruct-2501Mistral

33K$0.050$0.080—2026-10-27

Nnemotron-lightning-3.5-30b-a3b—

○—$0.050$0.200—Fireworks AI only2026-10-22

nvidia/nemotron-3-nano-30b-a3bNVIDIA

262K$0.050$0.200—2026-10-22

openai/gpt-4.1-nano:batchOpenAI

1M$0.050$0.200$0.0132026-10-27

openai/gpt-5-nanoOpenAI

400K$0.050$0.400$0.0052026-10-27

openai/gpt-6-lunaOpenAI

1.1M$0.050$0.250$0.0052026-10-20JSON

openai/gpt-6-luna-pro:batchOpenAI

1.1M$0.050$0.250$0.0052026-10-27

openai/gpt-6-luna:batchOpenAI

1.1M$0.050$0.250$0.0052026-10-22JSON

Ssao10k/llama-3-8b-stheno-v3.2Sao10k

○—$0.050$0.050—Novita AI only2026-10-22

Uupstage/solar-mini4Upstage

524K$0.050$0.200$0.005OpenRouter only2026-10-27

z-ai/glm-5.3-flash:batchZ.ai

1M$0.050$0.180$0.0102026-10-27

amazon/nova-lite-v1Amazon

300K$0.060$0.240—2026-10-27

ibm-granite/granite-4.2-8bIBM

131K$0.060$0.250$0.015OpenRouter only2026-10-27

inclusionai/ling-3.0-flash-vlInclusionAI

131K$0.060$0.180$0.0122026-10-27

poolside/laguna-xs-2.1Poolside

262K$0.060$0.120$0.030OpenRouter only2026-10-27

z-ai/glm-4.7-flashZ.ai

200K$0.060$0.400—2026-10-27

qwen/qwen3.5-flash-02-23Alibaba

1M$0.065$0.260—OpenRouter only2026-10-27

~z-ai/glm-latestZ.ai

1M$0.070$1.40$0.0692026-10-27

z-ai/glm-5.3Z.ai

1M$0.070$1.40$0.0692026-10-27

Bbaidu/ernie-4.5-21b-a3bBaidu

○—$0.070$0.280—Novita AI only2026-10-22

microsoft/phi-4Microsoft

16K$0.070$0.140—OpenRouter only2026-10-27

nvidia/nemotron-3.5-lightningNVIDIA

262K$0.070$0.200$0.040OpenRouter only2026-10-27

qwen/qwen3-coder-30b-a3b-instructAlibaba

262K$0.070$0.280—OpenRouter only2026-10-27

tencent/hy-mt2-30b-a3bTencent

8K$0.074$0.295—OpenRouter only2026-10-27

tencent/hy-mt2-7bTencent

8K$0.074$0.295—OpenRouter only2026-10-27

bytedance-seed/seed-1.6-flashByteDance

262K$0.075$0.300—OpenRouter only2026-10-27

inclusionai/ling-3.0-flash-finInclusionAI

262K$0.075$0.220$0.015Novita AI only2026-10-22

inclusionai/ling-3.1-flashInclusionAI

262K$0.075$0.225$0.0152026-10-27

mistralai/ministral-8b-2512:batchMistral

262K$0.075$0.075$0.0072026-10-26

mistralai/mistral-small-2603:batchMistral

262K$0.075$0.300$0.0072026-10-26

mistralai/mistral-small-3.2-24b-instructMistral

256K$0.075$0.200—2026-10-27

Nnex-agi/nex-n2.5-proNex-agi

262K$0.075$0.250$0.015OpenRouter only2026-10-27

openai/gpt-4o-mini:batchOpenAI

128K$0.075$0.300$0.0372026-10-27

openai/gpt-oss-safeguard-20bOpenAI

131K$0.075$0.300$0.0372026-10-27

Pprism-ml/ternary-bonsai-2-27bPrism-ml

262K$0.075$0.500$0.037OpenRouter only2026-10-27

google/gemma-3-27b-itGoogle

131K$0.080$0.350$0.0402026-10-27

Ggryphe/mythomax-l2-13bGryphe

8K$0.080$0.110—OpenRouter only2026-10-27

qwen/qwen3-32bAlibaba

131K$0.080$0.280—OpenRouter only2026-10-27

google/gemma-4-26b-a4b-itGoogle

262K$0.090$0.300$0.0502026-10-27

poolside/laguna-s-2.1Poolside

1M$0.090$0.180$0.009OpenRouter only2026-10-27

qwen/qwen3-235b-a22b-2507Alibaba

262K$0.090$0.550—OpenRouter only2026-10-27

qwen/qwen3-next-80b-a3b-instructAlibaba

262K$0.090$1.10$0.070OpenRouter only2026-10-27

~anthropic/claude-haiku-latestAnthropic

1M$0.100$0.500$0.0102026-10-27

~openai/gpt-luna-latestOpenAI

1.1M$0.100$0.500$0.0102026-10-27

anthropic/claude-haiku-5.5Anthropic

1M$0.100$0.500$0.0102026-10-27

bytedance-seed/seed-2.0-miniByteDance

262K$0.100$0.400—OpenRouter only2026-10-27

bytedance-seed/ui-tars-1.5-7bByteDance

128K$0.100$0.200$0.100OpenRouter only2026-10-27

google/gemini-2.5-flash-liteGoogle

1M$0.100$0.400$0.0102026-10-27

Iinception/mercury-3Inception

1M$0.100$0.350$0.010OpenRouter only2026-10-27

meta-llama/llama-3.3-70b-instructMeta

131K$0.100$0.320$0.2952026-10-27

meta-llama/llama-4-scoutMeta

1.3M$0.100$0.300—OpenRouter only2026-10-27

meta/muse-spark-1.2-contributorMeta

1M$0.100$0.200$0.002OpenRouter only2026-10-27

meta/muse-spark-1.3-contributorMeta

1M$0.100$0.200$0.002OpenRouter only2026-10-27

mistralai/ministral-3-3bMistral

○—$0.100$0.100$0.010Mistral AI only2026-10-22JSON

mistralai/ministral-3b-2512Mistral

131K$0.100$0.100$0.0102026-10-27

mistralai/voxtral-small-24b-2507Mistral

33K$0.100$0.300$0.0102026-10-27

openai/gpt-4.1-nanoOpenAI

1M$0.100$0.400$0.0302026-10-27

openai/gpt-5.4-nano:batchOpenAI

400K$0.100$0.625$0.0102026-10-27

openai/gpt-5.6-lunaOpenAI

1.1M$0.100$0.600$0.0102026-10-20JSON

openai/gpt-5.6-luna-pro:batchOpenAI

1.1M$0.100$0.600$0.0102026-10-27

openai/gpt-5.6-luna:batchOpenAI

1.1M$0.100$0.600$0.0102026-10-22JSON

openai/gpt-6-luna-proOpenAI

1.1M$0.100$0.500$0.0102026-10-27

openai/gpt-oss-120bOpenAI

131K$0.100$0.011$0.5002026-10-22

qwen/qwen-2.5-7b-instructAlibaba

33K$0.100$0.200—OpenRouter only2026-10-27

qwen/qwen3-30b-a3b-instruct-2507Alibaba

262K$0.100$0.300—OpenRouter only2026-10-27

qwen/qwen3.6-35b-a3bAlibaba

262K$0.100$0.950$0.0502026-10-27

Rrekaai/reka-edgeRekaai

16K$0.100$0.100—OpenRouter only2026-10-27

Rrekaai/reka-flash-3Rekaai

66K$0.100$0.200—OpenRouter only2026-10-27

stepfun/step-3.5-flashStepfun

262K$0.100$0.300—OpenRouter only2026-10-27

deepseek/deepseek-v4-flash-0731Deepseek

1M$0.119$0.238$0.0242026-10-27

google/gemma-3-27bGoogle

○—$0.119$0.200—2026-10-22

qwen/qwen3-14bAlibaba

41K$0.120$0.240—OpenRouter only2026-10-27

qwen/qwen3-30b-a3bAlibaba

41K$0.120$0.500—OpenRouter only2026-10-27

qwen/qwen3-coder-nextAlibaba

262K$0.120$0.800$0.070OpenRouter only2026-10-27

google/gemini-3.1-flash-lite:batchGoogle

1M$0.125$0.750$0.0132026-10-27

openai/gpt-5-mini:batchOpenAI

400K$0.125$1.00$0.0132026-10-27

google/gemma-4-26b-a4bGoogle

○—$0.130$0.400—2026-10-22

z-ai/glm-4.5-airZ.ai

131K$0.130$0.850$0.0252026-10-22

tencent/hy3Tencent

262K$0.132$0.528$0.0332026-10-27

deepseek/deepseek-v4.1-flashDeepseek

1M$0.140$0.420$0.0042026-10-27

google/gemma-4-31bGoogle

○—$0.140$0.400—2026-10-22

google/gemma-4-31b-itGoogle

262K$0.140$0.400—2026-10-21

qwen/qwen3.5-35b-a3bAlibaba

262K$0.140$1.00$0.0502026-10-27

tencent/hunyuan-a13b-instructTencent

131K$0.140$0.570—OpenRouter only2026-10-27

xiaomi/mimo-v2.6-flashXiaomi

1.1M$0.140$0.280$0.0032026-10-22

cohere/command-r-08-2024Cohere

128K$0.150$0.600—2026-10-27

deepseek/deepseek-v4.1-flash-1017Deepseek

1M$0.150$0.600$0.0032026-10-22

google/gemini-2.5-flash:batchGoogle

1M$0.150$1.25$0.0302026-10-20

google/gemini-3.5-flash-liteGoogle

1M$0.150$1.25—2026-10-22

google/gemini-3.5-flash-lite:batchGoogle

1M$0.150$1.25—2026-10-22

minimax/minimax-m3.1-flashMinimax

1M$0.150$0.600$0.0152026-10-27

mistralai/codestral-2508:batchMistral

256K$0.150$0.450$0.0152026-10-26

mistralai/ministral-3-8bMistral

○—$0.150$0.150$0.015Mistral AI only2026-10-22JSON

mistralai/ministral-8b-2512Mistral

262K$0.150$0.150$0.0152026-10-27

mistralai/mistral-small-2603Mistral

262K$0.150$0.600$0.0152026-10-27

mistralai/mistral-small-4Mistral

○—$0.150$0.600$0.015Mistral AI only2026-10-22JSON

nvidia/nemotron-3.5-super-vlNVIDIA

1M$0.150$0.500$0.0202026-10-27

openai/gpt-4o-miniOpenAI

128K$0.150$0.600$0.0752026-10-27

openai/gpt-4o-mini-2024-07-18OpenAI

128K$0.150$0.600$0.0752026-10-27

Pperceptron/perceptron-mk1Perceptron

33K$0.150$1.50—OpenRouter only2026-10-27

Pperceptron/perceptron-mk1.5Perceptron

37K$0.150$1.50—OpenRouter only2026-10-27

qwen/qwen3-next-80b-a3b-thinkingAlibaba

262K$0.150$1.20—OpenRouter only2026-10-27

qwen/qwen3-vl-30b-a3b-instructAlibaba

262K$0.150$0.600—2026-10-27

qwen/qwen3.8-flashAlibaba

1M$0.150$0.470$0.0162026-10-22

qwen/qwen3.8-omni-flashAlibaba

1M$0.150$0.470$0.016OpenRouter only2026-10-27

Rrouter-glm-5.3-flash—

○—$0.150$0.500—Fireworks AI only2026-10-22

Uupstage/solar-pro-3Upstage

131K$0.150$0.600$0.015OpenRouter only2026-10-27

Uupstage/solar-pro4Upstage

524K$0.150$0.600$0.030OpenRouter only2026-10-27

z-ai/glm-5.3-flashZ.ai

1M$0.150$0.500$0.0302026-10-22

xiaomi/mimo-v2.5Xiaomi

1.1M$0.168$0.336$0.0032026-10-22

qwen/qwen3.5-9bAlibaba

262K$0.170$0.250$0.040OpenRouter only2026-10-27

~moonshotai/kimi-latestMoonshot

1M$0.179$10.56$0.1702026-10-27

meta-llama/llama-guard-4-12bMeta

164K$0.180$0.180—OpenRouter only2026-10-27

meta/llama-4-scout-instructMeta

○—$0.180$0.590—Novita AI only2026-10-22

tencent/hy3-previewTencent

262K$0.180$0.600$0.060OpenRouter only2026-10-27

meta-llama/llama-4-maverickMeta

524K$0.188$0.652$0.050OpenRouter only2026-10-27

qwen/qwen3.6-flashAlibaba

1M$0.188$1.13—OpenRouter only2026-10-27

moonshotai/kimi-k3Moonshot

1M$0.189$11.50$0.1792026-10-27

qwen/qwen3-coder-flashAlibaba

1M$0.195$0.975$0.039OpenRouter only2026-10-27

Aai21/jamba-miniAi21

○—$0.200$0.400—AI21 Labs only2026-10-22

Ccognitivecomputations/dolphin-mistral-24b-venice-editionCognitivecomputations

128K$0.200$0.900—OpenRouter only2026-10-27

minimax/minimax-01Minimax

1M$0.200$1.10—OpenRouter only2026-10-27

mistralai/ministral-14b-2512Mistral

262K$0.200$0.200$0.0202026-10-27

mistralai/ministral-3-14bMistral

○—$0.200$0.200$0.020Mistral AI only2026-10-22JSON

mistralai/mistral-medium-3.1:batchMistral

131K$0.200$1.00$0.0202026-10-26

mistralai/mistral-sabaMistral

33K$0.200$0.600$0.0202026-10-27

nvidia/nemotron-3.5-content-safetyNVIDIA

131K$0.200$0.200—OpenRouter only2026-10-27

openai/gpt-4.1-mini:batchOpenAI

1M$0.200$0.800$0.0502026-10-27

openai/gpt-5.4-nanoOpenAI

400K$0.200$1.25$0.0202026-10-27

openai/gpt-5.6-luna-proOpenAI

1.1M$0.200$1.20$0.0202026-10-27

qwen/qwen3.8-27bAlibaba

1M$0.200$2.50$0.0902026-10-27

stepfun/step-3.7-flashStepfun

262K$0.200$1.15$0.0402026-10-22

qwen/qwen3-vl-235b-a22b-instructAlibaba

262K$0.210$1.90$0.100OpenRouter only2026-10-27

deepseek/deepseek-v4-flash-vision-expDeepseek

1M$0.216$0.647$0.0072026-10-27

qwen/qwen3.7-plus:batchAlibaba

262K$0.230$0.920—OpenRouter only2026-10-27

~deepseek/deepseek-pro-latestDeepseek

1M$0.243$8.00$0.2402026-10-27

Aarcee-ai/trinity-large-thinkingArcee-ai

262K$0.250$0.800$0.060OpenRouter only2026-10-26

bytedance-seed/seed-1.6ByteDance

262K$0.250$2.00—OpenRouter only2026-10-27

bytedance-seed/seed-2.0-liteByteDance

262K$0.250$2.00—OpenRouter only2026-10-27

deepseek/deepseek-chat-v3.1Deepseek

164K$0.250$0.950$0.1302026-10-27

google/gemini-2.5-flash-preview-tts:batchGoogle

○—$0.250$5.00—2026-10-22

google/gemini-3-flash-preview:batchGoogle

1M$0.250$1.50—2026-10-27

google/gemini-3.1-flash-liteGoogle

1M$0.250$1.50$0.0252026-10-27

google/gemini-3.1-flash-lite-imageGoogle

66K$0.250$1.50—2026-10-27

google/gemini-3.1-flash-lite-previewGoogle

1M$0.250$1.50$0.0252026-10-27

Iinception/mercury-2Inception

128K$0.250$0.750$0.025OpenRouter only2026-10-27

mistralai/mistral-large-2512:batchMistral

262K$0.250$0.750$0.0252026-10-26

openai/gpt-3.5-turbo:batchOpenAI

16K$0.250$0.750—2026-10-24

openai/gpt-5-miniOpenAI

400K$0.250$2.00$0.0252026-10-27

openai/gpt-5.1-codex-miniOpenAI

400K$0.250$2.00$0.0302026-10-27

qwen/qwen3-vl-8b-instructAlibaba

262K$0.250$0.750$0.120OpenRouter only2026-10-27

Rreflection/beam-1.0-501b-a23bReflection

1M$0.250$1.00$0.0502026-10-27

minimax/minimax-m2Minimax

205K$0.255$1.02—OpenRouter only2026-10-27

deepseek/deepseek-chatDeepseek

164K$0.257$1.03—2026-10-27

qwen/qwen-plusAlibaba

1M$0.260$0.780$0.052OpenRouter only2026-10-27

qwen/qwen3.5-122b-a10bAlibaba

262K$0.260$2.08—2026-10-27

qwen/qwen3.5-plus-02-15Alibaba

1M$0.260$1.56—OpenRouter only2026-10-27

deepseek/deepseek-v3.1-terminusDeepseek

164K$0.270$1.00—2026-10-27

deepseek/deepseek-v3.2-expDeepseek

164K$0.270$0.410—2026-10-27

meta/llama-4-maverick-instructMeta

○—$0.270$0.850—Novita AI only2026-10-22

Bbaidu/cobuddyBaidu

○—$0.280$1.13$0.070Novita AI only2026-10-22

deepseek/deepseek-v3.2Deepseek

164K$0.280$0.420$0.0282026-10-27

deepseek/deepseek-chat-v3-0324Deepseek

164K$0.290$1.14$0.1102026-10-27

qwen/qwen3-vl-30b-a3b-thinkingAlibaba

262K$0.290$1.00—OpenRouter only2026-10-27

amazon/nova-2-lite-v1Amazon

1M$0.300$2.50—2026-10-27

cohere/command-a-plusCohere

192K$0.300$1.50$0.1502026-10-27

google/gemini-2.5-flashGoogle

1M$0.300$2.50$0.0302026-10-27

Showing 200 of 509 matching · 509 priced in total · input and output are separate prices and are never averaged

Frequently asked

What is the cheapest LLM API right now?

ibm-granite/granite-4.0-h-micro at OpenRouter: $0.017 per million input tokens and $0.112 per million output, the lowest input rate among the 677 models we hold both halves for. Cheapest is not the same as suitable — a model that needs three attempts at your task costs more than one that needs a single attempt, and we hold no measurement of that.

Why one row per model instead of one per listing?

Because a seller listing the same model at three tiers is one seller, not three, and counting the tiers would let a single catalogue manufacture a spread on its own. Each seller appears once per model at its cheapest tier, and the Sellers column says how many stand behind the row.

What does “cached in” mean?

The rate for input tokens the seller has already seen — a system prompt or a document you send with every call. 356 of these models publish one. A seller with no cached tier bills a repeated prefix at the full input rate, which is why that column reads “none” rather than $0.00.

Why is the same model different prices at different sellers?

Because a seller is buying compute and reselling it, and they buy it on different terms. The weights are the same; the hardware under them, the batching, the margin and the contract are not. This is why the table is one row per model with its sellers underneath rather than a single price per model — there is no single price to print.

What is a batch tier?

A cheaper rate for work that does not need an answer now. Where a seller publishes one we show it in its own column, and it is read from their page rather than derived — a dash means that seller prices no batch tier, not that batch is free.

Are free models really free?

They are published at zero, which is not the same as unlimited. A free tier usually carries a rate limit or a preview status the seller can withdraw, and we record the price rather than the terms. The free listings are kept out of the cheapest ranking for that reason and are listed separately.

Can a seller pay to appear higher here?

No. The table has one stated sort at a time and nothing in it can be bought — a sponsor cannot enter it, reorder a row or change a rate. How the site is paid for is set out under money and independence at /our-data#sec5.

Estimate a workload

These are rates, not a bill. Put your own token mix in and the calculator prices it across every seller, cached share included.

Open the LLM cost calculator →

Every rate is what that seller charges to route to the model, not the model maker’s own price · readings over 21 days old are hidden rather than shown stale · no blended cost-per-quality figure appears here, because we hold no benchmark that would support one