#tHow Many Tokens?

← Back to counter

Cheapest AI model for batch summarization

Summarizing documents at scale is the workload most punished by input pricing. Long input (often 4-16k tokens), short output (200-800 tokens). We picked 8,000 input + 500 output as the midpoint and ranked every model on that shape.

GPT-5 Nano is the cheapest model for this workload at $600 per 1M calls (OpenAI). GPT-4.1 Nano is second at $1,000.

Ranked cheapest first

#ModelInput $/MOutput $/MPer 1M calls
#1 GPT-5 Nano
OpenAI
$0.05 $0.40 $600
#2 GPT-4.1 Nano
OpenAI
$0.10 $0.40 $1,000
#3 Gemini 2.5 Flash-Lite
Google
$0.10 $0.40 $1,000
#4 GPT-4o mini
OpenAI
$0.15 $0.60 $1,500
#5 Llama 3.1 8B
Meta
$0.18 $0.18 $1,530
#6 GPT-5.4 Nano
OpenAI
$0.20 $1.25 $2,225
#7 DeepSeek V3
DeepSeek
$0.27 $1.10 $2,710
#8 Gemini 3.1 Flash-Lite
Google
$0.25 $1.50 $2,750
#9 GPT-5 Mini
OpenAI
$0.25 $2.00 $3,000
#10 Gemini 2.5 Flash
Google
$0.30 $2.50 $3,650
#11 GPT-4.1 Mini
OpenAI
$0.40 $1.60 $4,000
#12 Llama 3.1 70B
Meta
$0.59 $0.79 $5,115
#13 Gemini 3 Flash
Google
$0.50 $3.00 $5,500
#14 DeepSeek V3.1
DeepSeek
$0.60 $1.70 $5,650
#15 Qwen 2.5 Coder 32B
Alibaba
$0.80 $0.80 $6,800
#16 Llama 3.3 70B
Meta
$0.88 $0.88 $7,480
#17 Qwen 2.5 72B
Alibaba
$0.90 $0.90 $7,650
#18 GPT-5.4 Mini
OpenAI
$0.75 $4.50 $8,250
#19 Claude Haiku 4.5
Anthropic
$1.00 $5.00 $10,500
#20 o3-mini
OpenAI
$1.10 $4.40 $11,000
#21 o4-mini
OpenAI
$1.10 $4.40 $11,000
#22 GLM-5.1
zhipu
$1.40 $4.40 $13,400
#23 GPT-5.1
OpenAI
$1.25 $10.00 $15,000
#24 GPT-5
OpenAI
$1.25 $10.00 $15,000
#25 Gemini 2.5 Pro
Google
$1.25 $10.00 $15,000
#26 Qwen3 Coder 480B
Alibaba
$2.00 $2.00 $17,000
#27 Mistral Large
Mistral
$2.00 $6.00 $19,000
#28 GPT-4.1
OpenAI
$2.00 $8.00 $20,000
#29 o3
OpenAI
$2.00 $8.00 $20,000
#30 GPT-5.3
OpenAI
$1.75 $14.00 $21,000
#31 GPT-5.2
OpenAI
$1.75 $14.00 $21,000
#32 Gemini 3.1 Pro
Google
$2.00 $12.00 $22,000
#33 GPT-4o
OpenAI
$2.50 $10.00 $25,000
#34 GPT-5.4
OpenAI
$2.50 $15.00 $27,500
#35 DeepSeek R1
DeepSeek
$3.00 $7.00 $27,500
#36 Llama 3.1 405B
Meta
$3.50 $3.50 $29,750
#37 Claude Sonnet 4.6
Anthropic
$3.00 $15.00 $31,500
#38 Claude Opus 4.8
Anthropic
$5.00 $25.00 $52,500
#39 GPT-5.5
OpenAI
$5.00 $30.00 $55,000
#40 GPT-4 Turbo
OpenAI
$10.00 $30.00 $95,000
#41 Claude Opus 4.8 (Fast Mode)
Anthropic
$10.00 $50.00 $105,000
#42 GPT-5 Pro
OpenAI
$15.00 $120.00 $180,000
#43 o3-pro
OpenAI
$20.00 $80.00 $200,000
#44 GPT-5.2 Pro
OpenAI
$21.00 $168.00 $252,000
#45 GPT-5.5 Pro
OpenAI
$30.00 $180.00 $330,000
#46 GPT-5.4 Pro
OpenAI
$30.00 $180.00 $330,000

Workload assumption: 8,000 input tokens + 500 output tokens per call, scaled to 1M calls. Pricing as of 2026-05-31.

How we computed this

At 8,000 input + 500 output, the input side is 16x the output side by volume, so input price is nearly the whole bill. An 8k-token document is roughly 10-12 pages of dense prose; if your documents run longer, chunk-then-merge summarization (summarize sections, then summarize the summaries) usually beats one giant call on both cost and quality. The other big lever is batch APIs: OpenAI, Anthropic, and Google all knock 50% off for asynchronous jobs with a 24-hour turnaround, and summarization pipelines are precisely the workload that tolerates that. The ranking below shows real-time pricing; halve the winner’s number if you can batch.

The math, worked through

One call at this workload costs GPT-5 Nano $0: 8,000 input tokens at $0.05 per million is $0, plus 500 output tokens at $0.40 per million is $0. At 10,000 calls a day that is $180 a month. The third-place model, Gemini 2.5 Flash-Lite, runs 1.7x that. The most expensive model in the table, GPT-5.4 Pro, costs 550x the winner at the same workload: the spread between top and bottom of this ranking is not a rounding error, it is the difference between a tool budget and a headcount budget.

About the winner

Budget models summarize fluently but compress lossily: they drop qualifiers, round numbers, and merge distinct claims. For news clipping or internal digests that is fine. For legal, medical, or financial document summarization, the cheapest model in this table is usually the wrong choice regardless of price.

When not to pick the cheapest

Watch two failure modes before scaling the winner: number scrambling (verify totals and dates survive the summary on a 50-document sample) and instruction drift on very long inputs (some models start ignoring format instructions past 6k input tokens). If either shows up, the next model up the price ladder is usually only 20-40% more expensive at this shape.

How to use this ranking

The winner is mathematically cheapest at the listed workload shape — that's not the same as "best for the use case." Cheaper models often have lower reasoning depth, smaller context windows, or worse instruction-following. Use this as the cost baseline, then test the top 2-3 candidates on your real prompts via the live counter.

Pricing snapshots come from each provider's published rate cards and are tracked in the full pricing changelog. Tokenizer accuracy per model is documented in the methodology.

Other ranked use cases

Try the live counter →