OpenAI API Pricing (October 2026)

OpenAI bills the API by the million tokens, in US dollars, with separate Standard, Batch, Flex, Fast and Ultrafast tables—not one flat rate for every request. On Standard short context (input tokens ≤272K), the current flagship line is gpt-6-astra at $10 input / $1 cached input / $50 output; gpt-6.1-sol at $2 / $0.10 / $10; gpt-6.1-terra at $1 / $0.05 / $5; gpt-6-sol at $2 / $0.20 / $10; and gpt-6-luna at $0.10 / $0.01 / $0.50. Responses, Chat Completions, Realtime, Batch and related endpoints share those model rates rather than charging a separate API-product fee. Figures below follow OpenAI's developers pricing documentation as of this write-up. For reference only. ChatGPT Plus and Pro seat plans are covered separately in ChatGPT Plus vs Pro—Price, Limits, and Who Should Upgrade (October 2026).

Flagship Standard rates (short context)

Short context means the request's input is ≤272K tokens. Long context is >272K input tokens, with higher per-million rates on the same model IDs. Prices are per 1M tokens. OpenAI also lists optional 30-minute and 12-hour prompt-cache write columns on many flagship rows; a cache read uses the cheaper cached-input cell once the write has been paid.

gpt-6-astra is the top Standard short-context row at $10 input, $1 cached input and $50 output. Cache writes on the same row are listed at $12.50 for a 30-minute write and $25 for a 12-hour write. Long context on astra is $20 / $2 / $75, with long cache writes at $25 and $50. In other words, crossing the 272K input threshold roughly doubles the input and cached-input cells on astra and lifts output from $50 to $75.

gpt-6.1-sol Standard short is $2 / $0.10 / $10, with 30-minute and 12-hour cache writes at $2.50 and $5. Long context is $4 / $0.20 / $15, with long cache writes at $5 and $10. That makes Sol 6.1 the mid-tier workhorse on the GPT-6 ladder when Astra-class reasoning is needed less often and Sol-class defaults most of the time.

gpt-6.1-terra Standard short is $1 / $0.05 / $5 (cache writes $1.25 / $2.50). Long context is $2 / $0.10 / $7.50 (long cache writes $2.50 / $5). Terra sits between Sol and Luna on the Standard short table and keeps the same short/long split as its siblings.

gpt-6-sol Standard short is $2 / $0.20 / $10 (cache writes $2.50 / $5). Long context is $4 / $0.40 / $15 (long cache writes $5 / $10). Note the cached-input gap versus gpt-6.1-sol: Sol 6.1 lists $0.10 cached short; Sol 6.0 lists $0.20. Input and output match between the two Sol IDs on Standard short; only the cache cell and the model generation differ.

gpt-6-luna is the low-cost flagship at $0.10 / $0.01 / $0.50 short (cache writes $0.125 / $0.25), and $0.20 / $0.02 / $0.75 long (long cache writes $0.25 / $0.50). Teams comparing Luna to older mini models should still verify token counts on the specific model ID, because cheaper rows can still dominate a bill when prompts are long or tools add system tokens.

The same Standard flagship table also lists the GPT-5.6 line beside the GPT-6 IDs: gpt-5.6-sol short $4 / $0.40 / $20; gpt-5.6-terra $2 / $0.20 / $12; gpt-5.6-luna $0.20 / $0.02 / $1.20. OpenAI states that GPT-5.6 Sol's promotional pricing remains available at least through November 21, 2026—after that window, use the live table rather than treating the promo as permanent. Long-context GPT-5.6 Sol is listed at $8 / $0.80 / $30 on Standard.

Batch and Flex

Batch and Flex publish separate tables. For the GPT-6 flagship short-context rows, both Batch and Flex match at half of Standard on the main input/output/cached columns: gpt-6-astra $5 / $0.50 / $25; gpt-6.1-sol $1 / $0.05 / $5; gpt-6.1-terra $0.50 / $0.025 / $2.50; gpt-6-sol $1 / $0.10 / $5; gpt-6-luna $0.05 / $0.005 / $0.25. Cache-write columns on those Batch/Flex rows are likewise half of the matching Standard short writes. Prefer the table for any model that is not on this flagship half-price pattern—pro rows and some older IDs follow their own Batch cells rather than a mental “always 50%” rule.

Flex is aimed at workloads that can tolerate variable latency in exchange for the lower table; Batch is the asynchronous bulk path. Neither changes the underlying model ID, and neither is a ChatGPT subscription discount.

Fast and Ultrafast

Fast is listed at about 2× Standard on the GPT-6 short-context flagships: astra $20 / $2 / $100; gpt-6.1-sol $4 / $0.20 / $20; terra $2 / $0.10 / $10; gpt-6-sol $4 / $0.40 / $20; luna $0.20 / $0.02 / $1.00. OpenAI notes that Priority processing was renamed Fast on July 30, 2026. Fast is a billing tier on the pricing page, not a published latency SLA, and it should not be read as a performance guarantee in marketing copy.

Ultrafast currently lists only gpt-6-astra and gpt-6.1-sol on the flagship Ultrafast table—not the full GPT-6 lineup. Short-context Ultrafast is $60 / $6 / $300 for astra and $12 / $0.60 / $60 for gpt-6.1-sol (about 6× Standard short). Long-context Ultrafast on those two IDs is $120 / $12 / $450 for astra and $24 / $1.20 / $90 for gpt-6.1-sol. Ultrafast rows for terra, Sol 6.0 or luna are not listed until the live table adds them.

Regional processing and FedRAMP

OpenAI applies a 10% uplift on regional processing (data residency) endpoints for models released on or after March 5, 2026 that are eligible for data residency, and the same 10% uplift on FedRAMP endpoints. Global default routing stays on the base table. Stack this uplift on top of whichever service tier (Standard / Batch / Flex / Fast / Ultrafast) is selected. A Fast regional request, for example, starts from the Fast cell and then adds the 10% residency uplift when that endpoint applies.

Older models still on the Standard list

The second Standard block keeps many GPT-5.x, GPT-4.x and o-series IDs. Representative short (or single-rate) rows include gpt-5.5 at $5 / $0.50 / $30 under the <272K labeling, gpt-5.5-pro at $30 / $180 (no cached column on that row), gpt-5.4 at $2.50 / $0.25 / $15, and gpt-5.4-pro at $30 / $180. Smaller IDs include gpt-5.4-mini $0.75 / $0.075 / $4.50, gpt-5.4-nano $0.20 / $0.02 / $1.25, gpt-5.2 $1.75 / $0.175 / $14, gpt-5.2-pro $21 / $168, gpt-5.1 and gpt-5 both $1.25 / $0.125 / $10, gpt-5-mini $0.25 / $0.025 / $2, gpt-5-nano $0.05 / $0.005 / $0.40, and gpt-5-pro $15 / $120.

Further down: gpt-4.1 $2 / $0.50 / $8, gpt-4.1-mini $0.40 / $0.10 / $1.60, gpt-4.1-nano $0.10 / $0.025 / $0.40, gpt-4o $2.50 / $1.25 / $10, gpt-4o-mini $0.15 / $0.075 / $0.60, o1 $15 / $7.50 / $60, o1-pro $150 / $600, o3-pro $20 / $80, o3 $2 / $0.50 / $8, o4-mini $1.10 / $0.275 / $4.40, o3-mini $1.10 / $0.55 / $4.40, and gpt-3.5-turbo $0.50 / $1.50. Batch halves many of these where a Batch row exists; Fast multiplies a subset. Always match the model ID and the service-tier tab before estimating a bill.

Realtime, image, embeddings and tools (brief)

Realtime and audio models are priced per 1M tokens by modality. On gpt-realtime-2.1, audio is $32 input / $0.40 cached / $64 output; text $4 / $0.40 / $24; image input $5 / $0.50 cached. Mini variants are lower (for example gpt-realtime-2.1-mini audio $10 / $0.30 / $20). GPT-Live 1 voice sessions are billed at $0.05 per minute without rounding up to a full minute, with backend model and tool usage charged separately.

Image generation lists gpt-image-2 / gpt-image-2.5 family rows around $8 image input / $2 cached / $30 image output and $5 text input on Standard, with Batch at half on the matching image rows. Embeddings remain cheap relative to chat: text-embedding-3-small $0.02 per 1M input tokens and text-embedding-3-large $0.13. omni-moderation-latest is free on the specialized table. Codex-oriented gpt-5.3-codex is $1.75 / $0.175 / $14 on Standard and $3.50 / $0.35 / $28 on Fast.

Built-in web search is commonly $10 per 1,000 calls plus search-content tokens at the chosen model's rates (preview non-reasoning variants list $25 / 1k with free search-content tokens on that preview row). File search storage is $0.10 / GB / day after 1 GB free; file-search tool calls on the Responses API are $2.50 / 1k. Hosted Shell and Code Interpreter containers are priced by size per 20-minute session (for example 1 GB $0.03, 4 GB $0.12). Tokens used inside tools still bill at the selected model rate. These lines matter for product cost models but are secondary to the chat flagship table above.

Short same-tier comparison

On Anthropic's current API list, Claude API Pricing (October 2026) puts Claude Sonnet 5.5 at $2 / $10 per million input/output tokens and Claude Haiku 5.5 from $0.10 / $0.50 for prompts up to 100K tokens—close to OpenAI's gpt-6.1-sol and gpt-6-luna Standard short rows in headline dollars, though tokenizers, cache multipliers and batch rules differ. DeepSeek's DeepSeek API Pricing (October 2026) lists deepseek-flash off-peak cache-miss input at $0.15 and output $0.60, with peak rates double those off-peak figures, so a like-for-like bill still depends on Beijing-time peak hours and cache hits. Equal dollars per million tokens are not equal dollars per task; compare on a fixed prompt after counting tokens on each vendor's model ID.

What this page is not

This article covers API token pricing for Responses, Chat Completions and related model endpoints on OpenAI's developers pricing documentation. It is not a ChatGPT Plus, Pro, Business or Enterprise seat guide—use the Plus vs Pro piece linked above for subscriptions. Service-tier availability can also vary by model; see OpenAI API Service Tier Restrictions when a tier is missing for an ID. Cyber and life-sciences specialty rows (for example gpt-5.6-cyber or Rosalind) sit on the same developers pricing page but are outside this guide's main chat comparison. Product details and tables change; the live OpenAI pricing documentation applies. For reference only.