LLM Cost Benchmark Dashboard
Batch workload benchmark for 300,000 records with 150 input tokens and 150 output tokens per record, balancing blended spend against latency, tail risk, and output-heavy pricing.
Models Benchmarked
–
–
Lowest Blended Cost
–
–
Fastest Median Latency
–
–
Flagged as Misleading
–
Decision framing
Bubble size = context window
Diamond markers = misleading unit-price signal
Log-scaled cost axes preserve low-end detail
Median and p99 included for batch SLA context
Sweet spot:
Pricing structure:
Long-tail spread:
Frontier models
–
Models that are not beaten on both cost and median latency at once.
Median benchmark point
–
Useful reference line for separating the lower-left cost/latency quadrant.
Highest blended run
–
Upper bound in the current benchmark set for this fixed batch workload.
Blended cost vs. median latency sweet spot
Total blended cost by model
Provider median benchmark profile
Input vs. output cost mix for the lowest-cost contenders
Median vs. p99 batch completion time for the cheapest shortlist
Top 10 lowest blended-cost models
| Rank | Model | Provider | Blended Cost | Median Latency | Median Batch Time | Flag |
|---|
Shortlist reading notes
The highlighted rows mark the current Pareto frontier, where no other benchmarked model is both cheaper and faster at the same time.
The stacked cost chart shows that output pricing quickly becomes the larger share of spend even when input rates look favorable, which is why the misleading flag matters for this equal-token workload.
The batch window chart translates per-request latency into a more operational question: how long the full 300,000-record job could take under median and p99 conditions.