GPT-5.4 API Pricing: Balanced Frontier Reasoning and Scale
Comprehensive GPT-5.4 API pricing analysis ($2.50/M input, $15.00/M output), reasoning efficiency, prompt caching ROI, and high-throughput production cost tiers.
How much does GPT-5.4 cost per million tokens?
GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens ($5.625/M blended at 3:1). Provides high-intelligence multi-step reasoning with cost predictability for enterprise production workloads. Verified 2026-09-08.
How much does GPT-5.4 cost per 1,000 requests?
Computed from generated token pricing. Each row assumes the listed input and output tokens per request; this model has no measured verbosity factor, so the unadjusted output estimate is shown.
| Request shape | Input tokens | Output tokens | Cost / 1,000 requests |
|---|---|---|---|
| Short | 100 | 50 | $1.0000 |
| Medium | 1,000 | 500 | $10.0000 |
| Long | 4,000 | 2,000 | $40.0000 |
Formula: ((input price × input tokens) + (output price × output tokens × verbosity factor)) ÷ 1,000,000 × 1,000. Assumptions: short 100/50, medium 1,000/500, long 4,000/2,000 input/output tokens per request. Unadjusted — no measured verbosity factor is available.
Three model-specific legacy pricing decisions
GPT-5.4 owns its dated base-model economics. GPT-5.6 Terra supplies the rate-parity check; GPT-5.6 Sol supplies the narrow successor crossover.
1. Fixed model-specific workload bills
| Workload | Input / output | 100K requests | Evidence boundary |
|---|---|---|---|
| Agent | 8,000 / 1,200 | $3800.00 | Text tokens |
| Long output | 32,000 / 4,000 | $14000.00 | Output sensitivity |
| Rate-parity check | 4,000 / 800 | $2200.00 | Same shape for Terra comparison |
Formula: requests × (input tokens × input $/M + output tokens × output $/M × output expansion) ÷ 1,000,000. Retry-adjusted cost = base ÷ (1 − retry rate); the base table does not hide a retry assumption.
2. GPT-5.4 → Sol accepted-result crossover
| Fixed shape | GPT-5.4 | GPT-5.6 Sol | Numeric decision boundary |
|---|---|---|---|
| Agent · 8,000 / 1,200 | $4130.43 | $5894.74 | 43% accepted-result uplift required after fixed retry assumptions (8% → 5%) |
| Long output · 32,000 / 4,000 | $15217.39 | $21894.74 | 43% accepted-result uplift required after fixed retry assumptions (8% → 5%) |
This is a cost-per-accepted-result threshold, not a measured quality claim. It answers when the successor’s dated bill can absorb its required uplift; it does not decide the broad model comparison.
3. Migration evidence matrix
| Traffic / evidence | Result | Safe treatment |
|---|---|---|
| 100K requests | $4130.43 | Fixed first workload; text-token units |
| 1M requests | $38000.00 | Linear token spend only; no volume discount inferred |
| 10M requests | $380000.00 | Budget exposure; quota and latency remain unavailable |
| Shutdown date: Unavailable; no current availability claim is made | Unavailable | Do not infer, zero-price, or import a neighboring model’s mechanic |
| Cache: Unavailable | Unavailable | Do not infer, zero-price, or import a neighboring model’s mechanic |
| Batch: Unavailable | Unavailable | Do not infer, zero-price, or import a neighboring model’s mechanic |
| Context and verbosity: Unavailable without model-specific dated evidence | Unavailable | Do not infer, zero-price, or import a neighboring model’s mechanic |
| Lifecycle | legacy; no sourced announcement date | No sourced shutdown date; revalidate before current claims |
Dated GPT-5.6 Terra rate-parity calculation
| Verified rate | GPT-5.4 | GPT-5.6 Terra | Delta | Conclusion |
|---|---|---|---|---|
| 2026-04-06 vs 2026-08-14 | $2.50 input / $15.00 output per 1M | $2.50 input / $15.00 output per 1M | $0.00 input / $0.00 output per 1M (0.0%) | Price alone cannot justify migration |
Parity calculation = Terra rate − GPT-5.4 rate, independently dated by the registry; both input and output deltas are zero, so migration needs non-price evidence.
Price verified 2026-04-06; lifecycle verified 2026-08-14. Luna is the data owner. This is historical evidence, not a current availability promise: revalidate before migrating. “Unavailable” means no compatible dated evidence was found; it is never treated as zero. First-party price source · lifecycle source · Test this model in All AI Ask.
Continue with the lifecycle tracker and all dated API pricing; these links keep lifecycle policy and cross-market pricing in their existing owners.
All three Batch 6 contributions are server-rendered for GPT-5.4; fixed inputs, formulas, dated provenance, successor boundary, lifecycle state, and missing-data treatment remain visible.
gpt-5-4GPT-5.4 API Pricing: Balanced Frontier Reasoning and Scale
GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens ($5.625/M blended at 3:1). Provides high-intelligence multi-step reasoning with cost predictability for enterprise production workloads. Verified 2026-09-08.
GPT-5.4 provides a direct cost-performance bridge between GPT-5.4 Mini and GPT-5.4 Pro, maintaining strict token bounds.
| Scenario | Rendered Evidence & Bounds |
|---|---|
| Scenario 1 | Light reasoning prompt (2K input, 500 output): $0.01250 total request cost |
| Scenario 2 | Document synthesis workload (16K input, 2K output): $0.07000 total request cost |
| Scenario 3 | Deep agentic multi-turn (64K input, 8K output): $0.28000 total request cost |
| Scenario 4 | Batch triage query (4K input, 250 output): $0.01375 total request cost |
| Scenario 5 | Enterprise code review payload (32K input, 4K output): $0.14000 total request cost |
| Scenario 6 | High-volume monthly run (250M tokens at 3:1): $1,406.25 monthly spend |
Prompt caching significantly lowers operational overhead for large agentic context windows.
| Scenario | Rendered Evidence & Bounds |
|---|---|
| Scenario 1 | System prompt prefix reuse (10K cached prefix, 2K delta): 36% prompt cost reduction |
| Scenario 2 | Large reference doc cached (50K cached, 5K delta): 45% prompt cost reduction |
| Scenario 3 | Codebase index cached (100K cached, 10K delta): 45% prompt cost reduction |
| Scenario 4 | Repeated customer chat context (4K cached, 1K delta): 40% prompt cost reduction |
| Scenario 5 | Workflow schema cache amortized over 20 turns: 47% cumulative input savings |
| Scenario 6 | Enterprise agent memory buffer (32K cached): $0.04000 savings per execution turn |
Batch API queuing allows high-volume asynchronous jobs to run at half the standard pricing rate.
| Scenario | Rendered Evidence & Bounds |
|---|---|
| Scenario 1 | Nightly compliance audit (50M input tokens): $62.50 batch cost vs $125.00 on-demand |
| Scenario 2 | Offline customer support evaluation (20M tokens): $56.25 batch cost vs $112.50 on-demand |
| Scenario 3 | Large corpus semantic categorization (100M tokens): $281.25 batch cost vs $562.50 on-demand |
| Scenario 4 | Code repository vulnerability scan (30M tokens): $84.38 batch cost vs $168.75 on-demand |
| Scenario 5 | Synthetic training data generation (80M tokens): $225.00 batch cost vs $450.00 on-demand |
| Scenario 6 | Enterprise asynchronous data lake processing: exactly 50% fixed infrastructure reduction |
How fast is GPT-5.4?
How much does GPT-5.4 cost at scale?
| Tokens / month | Est. cost (blended 3:1) |
|---|---|
| 100,000 | $0.56 |
| 1,000,000 | $5.63 |
| 10,000,000 | $56.25 |
| 100,000,000 | $562.50 |
How does GPT-5.4 compare with other models?
What should you explore next for GPT-5.4?
What are common questions about GPT-5.4?
Is GPT-5.4 cheaper than GPT-5.6 Terra?
GPT-5.4 costs $5.63/M blended tokens, GPT-5.6 Terra costs $5.63/M — GPT-5.6 Terra is cheaper.
How much does 1 million tokens cost with GPT-5.4?
At a 3:1 input:output ratio, 1 million blended tokens costs approximately $5.63. Pure input costs $2.50/M; pure output costs $15.00/M.
What does GPT-5.4 cost at high volume?
At 100 million blended tokens a month, GPT-5.4 costs approximately $562.50. See the cost-at-scale table below for other volumes.