OpenAIProprietaryReleased Mar 17, 2026GPT-5.4 Nano pricing, benchmarks and cost analysis
Epoch Capabilities Index (ECI)range 143.6 to 148.2
Throughput not measured for this model
In $0.200Out $1.25 per 1MCache discount 90%
Context window, tokens
Ranks are among the 101 models we track (intelligence = Epoch Capabilities Index by Epoch AI, CC BY 4.0; intelligence and speed: among those with data; cost: #1 is cheapest). Monthly cost uses the default workload of 2,000 input and 500 output tokens, 10,000 requests a day, 40% cached.
Comparison summary
GPT-5.4 Nano costs $264 a month for the default workload of 2,000 input and 500 output tokens, 10,000 requests a day with 40% cached input. That is 70% below the median model ($873). It ranks #24 of 101 on cost, where #1 is the cheapest.
It scores 146 on the Epoch Capabilities Index (#54 of 66). Among the 9 other models within 2 points, 1 costs less, and it costs 33% less than their median ($396). The cheapest stable-priced model that scores at least as high is Claude Haiku 5.5 (159.3) at $113 a month, 57% less.
Cached input is 90% cheaper than fresh input, in line with the 90% median. It is a proprietary model served by OpenAI. The batch API cuts the default workload to $132 a month.
Technical specifications
- Context window
- 400K tokens
- Input price
- $0.200 / 1M
- Cached input price
- $0.020 / 1M
- Output price
- $1.25 / 1M
- Cache discount
- 90%
- Batch API
- Available
- Batch input / cached / output
- $0.100 / $0.010 / $0.625
- Open weights
- No
- Release date
- Mar 17, 2026
- Price volatility
- Stable
Benchmarks
Best score for this model on each leaderboard, ranked against the other models listed there.
Your monthly bill
Standard
$264 / month- Input
- $72.00
- Cached input
- $4.80
- Output
- $187.50
Batch
$132 / month- Input
- $36.00
- Cached input
- $2.40
- Output
- $93.75
On the default workload the batch API costs $132 a month against $264 standard.
Price history
No price changes since we started tracking on Jul 13, 2026.
Cheaper alternatives
Models with stable pricing and a Epoch Capabilities Index within 2 points of GPT-5.4 Nano (or higher) that cost less for the default workload.
