Tools & price references
OpenAI API pricing
Check OpenAI API model prices, compare billing units and estimate your workload with selected text-model presets.
Free to useNo sign-upSources and formulas
Updated
Jump to the full model price reference
Estimated monthly cost
$80.00
USD / month10,000 requests · GPT-6.1 Sol
- Uncached input
- $30.00
- Cached input
- $0.00
- Output
- $50.00
- Other monthly costs
- $0.00
- API / request
- $0.0080
Standard input and output rates, no caching.
Estimate, not a quote. Excludes taxes, non-text usage, tool fees and charges you have not entered.
Official model price reference
88 named model entries across 112 pricing rows. Source checked . Calculator presets above remain a selected set.
All named model rows in the official pricing page Standard base-price sections checked on 2026-10-28, including modality rows and fine-tuning reference prices. Duplicate alternate processing modes and the duplicate GPT-5.6 Sol cyber listing are not repeated. Retired Sora aliases/snapshots are added only as unavailable reference entries from the official deprecation notice. This is not a list of every historical model or every account-accessible snapshot.
All reference rows shown
Text and reasoning: standard rates
USD per 1 million tokens. Standard processing. A dash means no rate listed, not free.
| Model / status | Input | Cached input | Output | Scope & source |
|---|---|---|---|---|
| gpt-6-astraCurrent | $10.00 | $1.00 | $50.00 | At up to 272,000 input tokens. Above 272,000, long-context input $20.00, cached input $2.00, output $75.00 per 1M tokens for the full request/session, not just the excess. Cache writes: 30m $12.50, 12h $25.00; long-context 30m $25.00, 12h $50.00 per 1M tokens.Official source |
| gpt-6.1-solCurrent | $2.00 | $0.10 | $10.00 | At up to 272,000 input tokens. Above 272,000, long-context input $4.00, cached input $0.20, output $15.00 per 1M tokens for the full request/session, not just the excess. Cache writes: 30m $2.50, 12h $5.00; long-context 30m $5.00, 12h $10.00 per 1M tokens.Official source |
| gpt-6.1-terraCurrent | $1.00 | $0.05 | $5.00 | At up to 272,000 input tokens. Above 272,000, long-context input $2.00, cached input $0.10, output $7.50 per 1M tokens for the full request/session, not just the excess. Cache writes: 30m $1.25, 12h $2.50; long-context 30m $2.50, 12h $5.00 per 1M tokens.Official source |
| gpt-6-lunaCurrent | $0.10 | $0.01 | $0.50 | At up to 272,000 input tokens. Above 272,000, long-context input $0.20, cached input $0.02, output $0.75 per 1M tokens for the full request/session, not just the excess. Cache writes: 30m $0.125, 12h $0.25; long-context 30m $0.25, 12h $0.50 per 1M tokens.Official source |
| gpt-6-solCurrent | $2.00 | $0.20 | $10.00 | At up to 272,000 input tokens. Above 272,000, long-context input $4.00, cached input $0.40, output $15.00 per 1M tokens for the full request/session, not just the excess. Cache writes: 30m $2.50, 12h $5.00; long-context 30m $5.00, 12h $10.00 per 1M tokens.Official source |
| gpt-5.6-solCurrent | $4.00 | $0.40 | $20.00 | At up to 272,000 input tokens. Above 272,000, long-context input $8.00, cached input $0.80, output $30.00 per 1M tokens for the full request/session, not just the excess. Cache writes: 30m $5.00, 12h $10.00; long-context 30m $10.00, 12h $20.00 per 1M tokens. Promotional rate available at least through 2026-11-21. Also listed under Cyber with identical rates.Official source |
| gpt-5.6-terraCurrent | $2.00 | $0.20 | $12.00 | At up to 272,000 input tokens. Above 272,000, long-context input $4.00, cached input $0.40, output $18.00 per 1M tokens for the full request/session, not just the excess. Cache writes: 30m $2.50, 12h $5.00; long-context 30m $5.00, 12h $10.00 per 1M tokens.Official source |
| gpt-5.6-lunaCurrent | $0.20 | $0.02 | $1.20 | At up to 272,000 input tokens. Above 272,000, long-context input $0.40, cached input $0.04, output $1.80 per 1M tokens for the full request/session, not just the excess. Cache writes: 30m $0.25, 12h $0.50; long-context 30m $0.50, 12h $1.00 per 1M tokens.Official source |
| gpt-5.5Current | $5.00 | $0.50 | $30.00 | At up to 272,000 input tokens. Above 272,000, long-context input $10.00, cached input $1.00, output $45.00 per 1M tokens for the full request/session, not just the excess.Official source |
| gpt-5.5-proCurrent | $30.00 | - | $180.00 | At up to 272,000 input tokens. Above 272,000, long-context input $60.00, cached input -, output $270.00 per 1M tokens for the full request/session, not just the excess.Official source |
| gpt-5.4Current | $2.50 | $0.25 | $15.00 | At up to 272,000 input tokens. Above 272,000, long-context input $5.00, cached input $0.50, output $22.50 per 1M tokens for the full request/session, not just the excess.Official source |
| gpt-5.4-proCurrent | $30.00 | - | $180.00 | At up to 272,000 input tokens. Above 272,000, long-context input $60.00, cached input -, output $270.00 per 1M tokens for the full request/session, not just the excess.Official source |
| gpt-5.4-miniCurrent | $0.75 | $0.075 | $4.50 | Official source |
| gpt-5.4-nanoDeprecated | $0.20 | $0.02 | $1.25 | Scheduled shutdown 2027-04-01.Official source |
| gpt-5.2Older generation | $1.75 | $0.175 | $14.00 | Official source |
| gpt-5.2-proOlder generation | $21.00 | - | $168.00 | Official source |
| gpt-5.1Deprecated | $1.25 | $0.125 | $10.00 | Scheduled shutdown 2027-04-01.Official source |
| gpt-5Deprecated | $1.25 | $0.125 | $10.00 | Scheduled shutdown 2026-12-11.Official source |
| gpt-5-miniDeprecated | $0.25 | $0.025 | $2.00 | Scheduled shutdown 2026-12-11.Official source |
| gpt-5-nanoDeprecated | $0.05 | $0.005 | $0.40 | Scheduled shutdown 2026-12-11.Official source |
| gpt-5-proDeprecated | $15.00 | - | $120.00 | Scheduled shutdown 2026-12-11.Official source |
| gpt-4.1Deprecated | $2.00 | $0.50 | $8.00 | Scheduled shutdown 2027-01-14.Official source |
| gpt-4.1-miniDeprecated | $0.40 | $0.10 | $1.60 | Scheduled shutdown 2027-01-14.Official source |
| gpt-4oDeprecated | $2.50 | $1.25 | $10.00 | Scheduled shutdown 2027-01-14.Official source |
| gpt-4o-miniDeprecated | $0.15 | $0.075 | $0.60 | Scheduled shutdown 2027-01-14.Official source |
| o3-proDeprecated | $20.00 | - | $80.00 | Scheduled shutdown 2026-12-11.Official source |
| o3Deprecated | $2.00 | $0.50 | $8.00 | Scheduled shutdown 2026-12-11.Official source |
Cybersecurity models
USD per 1 million tokens. Standard processing. A dash means no rate listed, not free.
| Model / status | Input | Cached input | Output | Scope & source |
|---|---|---|---|---|
| gpt-5.6-cyberLimited availability | $12.50 | $1.25 | $75.00 | Daybreak cyber listing. Access is restricted; confirm approval and provisioning. Long-context rates are not listed in the main pricing table. Maximum input 272,000 tokens in the model card. Cache writes: 30m $15.625, 12h $31.25 per 1M tokens. The model card also contains a generic >272K surcharge note despite that input cap; no unsupported long tier is inferred.Official source |
| gpt-5.5-cyberLimited availability | $12.50 | $1.25 | $75.00 | Daybreak cyber listing. Access is restricted; confirm approval and provisioning. Long-context rates are not listed in the main pricing table. Dedicated gpt-5.5-cyber model page was unavailable at verification; the main official pricing table remains its price evidence.Official source |
Life sciences models
USD per 1 million tokens. Standard processing. A dash means no rate listed, not free.
| Model / status | Input | Cached input | Output | Scope & source |
|---|---|---|---|---|
| gpt-rosalind-researchLimited availability | $5.00 | $0.50 | $25.00 | Limited to approved internal research through the trusted-access program. Billing began 2026-10-05; cache-write pricing does not apply.Official source |
| gpt-rosalind-discoveryLimited availability | $5.00 | $0.50 | $25.00 | Limited to approved internal research through the trusted-access program.Official source |
Live voice sessions
USD per minute; billed per second.
| Model / status | Session rate | Scope & source |
|---|---|---|
| gpt-live-1Current | $0.05 / minute | No rounding up to a whole minute. Backend model and tool usage billed separately.Official source |
Realtime and audio generation
USD per 1M tokens of the stated modality, except TTS character rates explicitly shown.
| Model / status | Modality | Input | Cached input | Output | Scope & source |
|---|---|---|---|---|---|
| gpt-realtime-2.1Current | Audio | $32.00 | $0.40 | $64.00 | Rates apply to the indicated modality.Official source |
| gpt-realtime-2.1Current | Text | $4.00 | $0.40 | $24.00 | Rates apply to the indicated modality.Official source |
| gpt-realtime-2.1Current | Image | $5.00 | $0.50 | - | Rates apply to the indicated modality.Official source |
| gpt-realtime-2.1-miniCurrent | Audio | $10.00 | $0.30 | $20.00 | Rates apply to the indicated modality.Official source |
| gpt-realtime-2.1-miniCurrent | Text | $0.60 | $0.06 | $2.40 | Rates apply to the indicated modality.Official source |
| gpt-realtime-2.1-miniCurrent | Image | $0.80 | $0.08 | - | Rates apply to the indicated modality.Official source |
| gpt-realtime-2Current | Audio | $32.00 | $0.40 | $64.00 | Rates apply to the indicated modality.Official source |
| gpt-realtime-2Current | Text | $4.00 | $0.40 | $24.00 | Rates apply to the indicated modality.Official source |
| gpt-realtime-2Current | Image | $5.00 | $0.50 | - | Rates apply to the indicated modality.Official source |
| gpt-realtime-1.5Current | Audio | $32.00 | $0.40 | $64.00 | Rates apply to the indicated modality.Official source |
| gpt-realtime-1.5Current | Text | $4.00 | $0.40 | $16.00 | Rates apply to the indicated modality.Official source |
| gpt-realtime-1.5Current | Image | $5.00 | $0.50 | - | Rates apply to the indicated modality.Official source |
| gpt-realtime-miniDeprecated | Audio | $10.00 | $0.30 | $20.00 | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-realtime-miniDeprecated | Text | $0.60 | $0.06 | $2.40 | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-realtime-miniDeprecated | Image | $0.80 | $0.08 | - | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-realtimeDeprecated | Audio | $32.00 | $0.40 | $64.00 | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-realtimeDeprecated | Text | $4.00 | $0.40 | $16.00 | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-realtimeDeprecated | Image | $5.00 | $0.50 | - | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-audio-1.5Current | Audio | $32.00 | - | $64.00 | Rates apply to the indicated modality.Official source |
| gpt-audio-1.5Current | Text | $2.50 | - | $10.00 | Rates apply to the indicated modality.Official source |
| gpt-audio-miniDeprecated | Audio | $10.00 | - | $20.00 | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-audio-miniDeprecated | Text | $0.60 | - | $2.40 | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-audioDeprecated | Audio | $32.00 | - | $64.00 | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-audioDeprecated | Text | $2.50 | - | $10.00 | Scheduled shutdown 2027-01-20. Rates apply to the indicated modality.Official source |
| gpt-4o-mini-ttsDeprecated | Audio | - | - | $12.00 | Scheduled shutdown 2027-01-06. Rates apply to the indicated modality.Official source |
| gpt-4o-mini-ttsDeprecated | Text | $0.60 | - | - | Scheduled shutdown 2027-01-06. Rates apply to the indicated modality.Official source |
| tts-1Deprecated | Text | $15.00 / 1M characters | - | - | Scheduled shutdown 2027-01-06. Character-based billing.Official source |
| tts-1-hdDeprecated | Text | $30.00 / 1M characters | - | - | Scheduled shutdown 2027-01-06. Character-based billing.Official source |
Image generation: standard rates
USD per 1M tokens of the stated modality; not a price per image.
| Model / status | Modality | Input | Cached input | Output | Scope & source |
|---|---|---|---|---|---|
| gpt-image-2.5-sunburstCurrent | Image | $8.00 | $2.00 | $30.00 | Cached rates apply only to images generated through the Responses API.Official source |
| gpt-image-2.5-sunburstCurrent | Text | $5.00 | $1.25 | - | Cached rates apply only to images generated through the Responses API.Official source |
| gpt-image-2.5-flareCurrent | Image | $8.00 | $2.00 | $30.00 | Cached rates apply only to images generated through the Responses API.Official source |
| gpt-image-2.5-flareCurrent | Text | $5.00 | $1.25 | - | Cached rates apply only to images generated through the Responses API.Official source |
| gpt-image-2Current | Image | $8.00 | $2.00 | $30.00 | Cached rates apply only to images generated through the Responses API.Official source |
| gpt-image-2Current | Text | $5.00 | $1.25 | - | Cached rates apply only to images generated through the Responses API.Official source |
| gpt-image-1.5Deprecated | Image | $8.00 | $2.00 | $32.00 | Scheduled shutdown 2026-12-01. Final image cost depends on token usage, size and quality.Official source |
| gpt-image-1.5Deprecated | Text | $5.00 | $1.25 | $10.00 | Scheduled shutdown 2026-12-01. Final image cost depends on token usage, size and quality.Official source |
| gpt-image-1-miniDeprecated | Image | $2.50 | $0.25 | $8.00 | Scheduled shutdown 2026-12-01. Final image cost depends on token usage, size and quality.Official source |
| gpt-image-1-miniDeprecated | Text | $2.00 | $0.20 | - | Scheduled shutdown 2026-12-01. Final image cost depends on token usage, size and quality.Official source |
| chatgpt-image-latestDeprecated | Image | $8.00 | $2.00 | $32.00 | Scheduled shutdown 2026-12-01. Final image cost depends on token usage, size and quality.Official source |
| chatgpt-image-latestDeprecated | Text | $5.00 | $1.25 | $10.00 | Scheduled shutdown 2026-12-01. Final image cost depends on token usage, size and quality.Official source |
Transcription and live translation
Input/output: USD per 1M tokens as published. Estimated costs: USD per minute.
| Model / status | Use case | Input | Output | Estimated cost | Scope & source |
|---|---|---|---|---|---|
| gpt-realtime-translateCurrent | Live translation | - | - | $0.034 / minute | Per-minute figures are the official estimated costs; do not add them to token charges.Official source |
| gpt-live-transcribeCurrent | Live transcription | - | - | $0.017 / minute | Per-minute figures are the official estimated costs; do not add them to token charges.Official source |
| gpt-realtime-whisperCurrent | Live transcription | - | - | $0.017 / minute | Per-minute figures are the official estimated costs; do not add them to token charges.Official source |
| gpt-transcribeCurrent | Transcription | - | - | $0.0045 / minute | Per-minute figures are the official estimated costs; do not add them to token charges.Official source |
| gpt-4o-transcribeDeprecated | Transcription | $2.50 | $10.00 | $0.006 / minute | Scheduled shutdown 2027-02-26. Per-minute figures are the official estimated costs; do not add them to token charges.Official source |
| gpt-4o-mini-transcribeDeprecated | Transcription | $1.25 | $5.00 | $0.003 / minute | Scheduled shutdown 2027-02-26. Per-minute figures are the official estimated costs; do not add them to token charges.Official source |
| gpt-4o-transcribe-diarizeDeprecated | Transcription + diarization | $2.50 | $10.00 | $0.006 / minute | Scheduled shutdown 2027-02-26. Per-minute figures are the official estimated costs; do not add them to token charges.Official source |
| whisper-1Deprecated | Transcription | - | - | $0.006 / minute | Scheduled shutdown 2027-02-26. Official pricing label: Whisper.Official source |
Chat, coding and search
USD per 1 million tokens. Standard processing. A dash means no rate listed, not free.
| Model / status | Input | Cached input | Output | Scope & source |
|---|---|---|---|---|
| chat-latestCurrent | $5.00 | $0.50 | $30.00 | ChatGPT.Official source |
| gpt-5.3-codexDeprecated | $1.75 | $0.175 | $14.00 | Scheduled shutdown 2027-04-01. Codex.Official source |
| gpt-5-search-apiCurrent | $1.25 | $0.125 | $10.00 | Search. Search tool charges may apply separately. Dedicated model page was unavailable; rate verified against main official pricing.Official source |
Embeddings
USD per 1M input tokens.
| Model / status | Input | Scope & source |
|---|---|---|
| text-embedding-3-smallCurrent | $0.02 | Official source |
| text-embedding-3-largeCurrent | $0.13 | Official source |
| text-embedding-ada-002Older generation | $0.10 | Official source |
Moderation
Official API moderation rate.
| Model / status | Rate | Scope & source |
|---|---|---|
| omni-moderation-latestCurrent | Free | Official source |
Retired text models: reference prices only
USD per 1 million tokens. Standard processing. A dash means no rate listed, not free. These models have already shut down.
| Model / status | Input | Cached input | Output | Scope & source |
|---|---|---|---|---|
| gpt-4.1-nanoRetired | $0.10 | $0.025 | $0.40 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| gpt-4o-2024-05-13Retired | $5.00 | - | $15.00 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| o1Retired | $15.00 | $7.50 | $60.00 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| o1-proRetired | $150.00 | - | $600.00 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| o4-miniRetired | $1.10 | $0.275 | $4.40 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| o3-miniRetired | $1.10 | $0.55 | $4.40 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| gpt-4-turbo-2024-04-09Retired | $10.00 | - | $30.00 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| gpt-4-0613Retired | $30.00 | - | $60.00 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| gpt-3.5-turboRetired | $0.50 | - | $1.50 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
| gpt-3.5-turbo-0125Retired | $0.50 | - | $1.50 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use.Official source |
Fine-tuning: restricted and legacy reference
USD per 1M tokens except training explicitly billed per hour. Standard inference rates.
| Model / status | Training | Input | Cached input | Output | Scope & source |
|---|---|---|---|---|---|
| o4-mini-2025-04-16Retired | $100.00 / hour | $4.00 | $1.00 | $16.00 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference.Official source |
| o4-mini-2025-04-16Retired | $100.00 / hour | $2.00 | $0.50 | $8.00 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference. Data-sharing inference discount shown.Official source |
| gpt-4.1-2025-04-14Deprecated | $25.00 | $3.00 | $0.75 | $12.00 | Scheduled shutdown 2027-01-14. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference.Official source |
| gpt-4.1-mini-2025-04-14Deprecated | $5.00 | $0.80 | $0.20 | $3.20 | Scheduled shutdown 2027-01-14. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference.Official source |
| gpt-4.1-nano-2025-04-14Retired | $1.50 | $0.20 | $0.05 | $0.80 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference.Official source |
| gpt-4o-2024-08-06Deprecated | $25.00 | $3.75 | $1.875 | $15.00 | Scheduled shutdown 2027-01-14. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference.Official source |
| gpt-4o-mini-2024-07-18Deprecated | $3.00 | $0.30 | $0.15 | $1.20 | Scheduled shutdown 2027-01-14. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference.Official source |
| gpt-3.5-turboRetired | $8.00 | $3.00 | - | $6.00 | Shut down 2026-10-23. Retained official reference price; unavailable for new API use. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference. Listed as legacy in the pricing table.Official source |
| davinci-002Retired | $6.00 | $12.00 | - | $12.00 | Shut down 2026-09-28. Retained official reference price; unavailable for new API use. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference. Listed as legacy in the pricing table.Official source |
| babbage-002Retired | $0.40 | $1.60 | - | $1.60 | Shut down 2026-09-28. Retained official reference price; unavailable for new API use. Self-serve fine-tuning is winding down. Eligible existing customers only; new jobs end 2027-01-06. Underlying model shutdown dates govern inference. Listed as legacy in the pricing table.Official source |
Video: retired, no current rate
No current video pricing is listed on the official pricing page.
| Model / status | Current price | Scope & source |
|---|---|---|
| sora-2Retired | Unavailable | Videos API and these models shut down 2026-09-24; no current rate is inferred.Official source |
| sora-2-proRetired | Unavailable | Videos API and these models shut down 2026-09-24; no current rate is inferred.Official source |
| sora-2-2025-10-06Retired | Unavailable | Videos API and these models shut down 2026-09-24; no current rate is inferred.Official source |
| sora-2-2025-12-08Retired | Unavailable | Videos API and these models shut down 2026-09-24; no current rate is inferred.Official source |
| sora-2-pro-2025-10-06Retired | Unavailable | Videos API and these models shut down 2026-09-24; no current rate is inferred.Official source |
A retained price row does not establish current API availability. Older, deprecated and retired models are labelled separately. Units, prompt tiers and access conditions are part of each price; consult the linked source before budgeting.
Calculator preset billing notes
The calculator uses the four selected text-model presets shown in its menu. The full reference above also includes models with other units, access conditions or retired status.
Higher prompt tiers
| Model | Threshold per request | Input / 1M | Output / 1M |
|---|---|---|---|
| GPT-6.1 Sol | Over 272,000 input tokens | $4 | $15 |
| GPT-6.1 Terra | Over 272,000 input tokens | $2 | $7.5 |
| GPT-6 Luna | Over 272,000 input tokens | $0.2 | $0.75 |
| GPT-6 Astra | Over 272,000 input tokens | $20 | $75 |
API usage and ChatGPT subscriptions
A ChatGPT subscription and API usage are different purchases. Enter your billed API usage here; a consumer subscription price is not an API token allowance.
Long prompts change the whole request
For the selected models, the recorded higher tier starts above 272,000 input tokens in a request. The calculator applies that tier before multiplying by your request volume. Billed reasoning belongs in the output token count.
Choose on cost and task quality
Use the same representative task across candidate models. Compare valid outputs, retries and review time alongside token cost. A lower per-token rate does not establish a lower cost per accepted result.
Cache-write reference: base prompt tier
- GPT-6.1 Sol: 30-minute cache write $2.5 / 1M tokens; 720-minute cache write $5 / 1M tokens
- GPT-6.1 Terra: 30-minute cache write $1.25 / 1M tokens; 720-minute cache write $2.5 / 1M tokens
- GPT-6 Luna: 30-minute cache write $0.125 / 1M tokens; 720-minute cache write $0.25 / 1M tokens
- GPT-6 Astra: 30-minute cache write $12.5 / 1M tokens; 720-minute cache write $25 / 1M tokens
The cached-input share models reads. Newly written tokens remain in ordinary input usage. To avoid counting their input charge twice, add only the write premium above that already-counted input charge, plus full storage costs, under other monthly costs. For example, 1M Sonnet input tokens at $2 with a $2.50 write rate need a $0.50 adjustment, not another $2.50. These reference rates use the base prompt tier; check the source for long-context write rates.