Cheapest GPT-5.4 lane for simple routing, extraction, and bulk automationGPT-5.4 nano by openai is offered by 38 providers on this page. Public prices are shown for 36 of them. Its official list price is $0.20 per 1M input tokens and $1.25 per 1M output tokens. Prices vary about 1.2× across the providers that host it. It also has 1 free ($0) channel; free tiers usually carry rate limits, and subscription-covered access bills $0 per token only after the subscription fee.
Median output speed is 161 tokens per second, with 7.55s to the first token.
The context window is 400,000 tokens at the reference host, but hosts report different limits, from 128,000 to 400,000, so the usable window depends on the provider you pick. It supports reasoning, tool use, and structured output. Accepted input modalities are Text, Image, and PDF. Its training knowledge cuts off at 2025-08-31.
Quality is independently evaluated by Artificial Analysis. Speed/latency are model-level medians.
| Provider | Tier | Input | Output | Cache read | Cache write | Context | Output limit | Status |
|---|---|---|---|---|---|---|---|---|
| GitLab Duo duo-chat-gpt-5-4-nano | Cloud | Free | — | — | 400,000 | 128,000 | deprecated | |
| Poe | Gateway | $0.18 | $1.14 | $0.018 | — | 400,000 | 128,000 | |
| OpenAIOfficial | First-party | $0.20 | $1.25 | $0.02 | — | 400,000 | 128,000 | |
| CrossModel | Gateway | $0.20 | $1.25 | $0.02 | $0.20 | 400,000 | 128,000 | |
| LLM Gateway | Gateway | $0.20 | $1.25 | $0.02 | — | 400,000 | 128,000 | |
| LLM Gateway | Gateway | $0.20 | $1.25 | $0.02 | — | 400,000 | 128,000 | |
| Azure Cognitive Services | Cloud | $0.20 | $1.25 | $0.02 | — | 400,000 | 128,000 | |
| 302.AI | Gateway | $0.20 | $1.25 | $0.02 | — | 400,000 | 128,000 | |
| Requesty | Gateway | $0.20 | $1.25 | $0.02 | — | 400,000 | 128,000 | |
| DigitalOcean openai-gpt-5.4-nano | Cloud | $0.20 | $1.25 | $0.02 | — | 400,000 | 128,000 |
Sorted by blended price (input×0.75 + output×0.25) asc. The official channel always shows regardless of rank. Whether a gateway's low price is actually usable can't be verified.
Interleaved thinking (reasoning between tool calls) is declared by 1 of 38 providers.
1 / 38 providers expose no reasoning control (reasoning_options: []).
| Mode | Provider | Input | Output | Cache read | Cache write |
|---|---|---|---|---|---|
| flex | OpenAI | $0.10 | $0.63 | $0.01 | — |
1 more channels offer $0 (GitLab Duo); free tiers usually have rate limits and no SLA, excluded from ranking.
| Name | Conditions | Score | Metric | Source |
|---|---|---|---|---|
| SWE-Bench Pro | variant: reasoning effort xhigh | 52.4 | resolve rate | Source ↗ |
| Terminal-Bench | variant: reasoning effort xhigh · v2.0 | 46.3 | accuracy | Source ↗ |
| MCP Atlas | variant: reasoning effort xhigh | 56.1 | score | Source ↗ |
| Toolathlon | variant: reasoning effort xhigh | 35.5 | score | Source ↗ |
| τ²-Bench Telecom | variant: reasoning effort xhigh | 92.5 | accuracy | Source ↗ |
| GPQA Diamond | variant: reasoning effort xhigh | 82.8 | accuracy | Source ↗ |
| Humanity's Last Exam | variant: with tools | 37.7 | accuracy | Source ↗ |
| Humanity's Last Exam | variant: without tools | 24.3 | accuracy | Source ↗ |
| OSWorld-Verified | variant: reasoning effort xhigh | 39 | success rate | Source ↗ |
| MMMU Pro | variant: with Python | 69.5 | accuracy | Source ↗ |
| MMMU Pro | variant: without tools | 66.1 | accuracy | Source ↗ |
| OmniDocBench | variant: reasoning effort none · v1.5 | 0.2419 | overall edit distance | Source ↗ |
| OpenAI MRCR | variant: 8-needle, 64K-128K · vv2 | 44.2 | accuracy | Source ↗ |
| OpenAI MRCR | variant: 8-needle, 128K-256K · vv2 | 33.1 | accuracy | Source ↗ |
| Graphwalks | variant: BFS, 0-128K | 73.4 | accuracy | Source ↗ |
| Graphwalks | variant: parents, 0-128K | 50.8 | accuracy | Source ↗ |
The same benchmark scores very differently across harness / dataset, so the qualifying conditions must be shown together.