OpenAI API Pricing
The lab behind ChatGPT and the GPT family. Founded as a nonprofit in 2015 by Sam Altman, Elon Musk, Ilya Sutskever, Greg Brockman and others; restructured into a capped-profit LLC in 2019 and finally into a public-benefit corporation on October 28, 2025. Ships the largest commercial model lineup, the largest consumer AI product, and — after the April 2026 funding round — the highest-valued private AI company in history at ~$852B.
GPT-6.1 Sol LAUNCHED SEP 2026
GPT-6.1 Sol is OpenAI's balanced GPT-6 model, released 2026-09-29 and described on its model page as delivering “near-Astra performance at a lower cost for complex coding, computer use, and profess…
GPT-6 Sol LAUNCHED SEP 2026
GPT-6 Sol is the Sol OpenAI released on 2026-09-22, still on the price list after GPT-6.1 Sol took its place in OpenAI's lineup on 2026-09-29 with the same input and output rates and half the cache…
GPT-6 Luna LAUNCHED SEP 2026
GPT-6 Luna is OpenAI's high-volume tier, released 2026-09-22 at $0.10/M input, $0.01/M cached input, $0.125/M cache writes and $0.50/M output - half the $0.20/$1.20 of GPT-5.6 Luna.
GPT-6 Astra LAUNCHED SEP 2026
GPT-6 Astra is OpenAI's new top-of-stack model, announced 2026-09-03 and described in the docs as “our most capable model, built for the hardest end-to-end work”; models.md now opens with “if you'r…
GPT-5.6 Cyber
⚠ Daybreak programme only - separate approval and provisioning for approved defenders doing authorised vulnerability research. Apply at openai.com/daybreak. OpenAI's cybersecurity model for approve…
GPT-5.6 Sol
GPT-5.6 Sol is OpenAI's frontier GPT-5.6 tier, generally available since 2026-07-09.
GPT-5.6 Terra
GPT-5.6 Terra is OpenAI's balanced GPT-5.6 tier, generally available since 2026-07-09.
GPT-5.6 Luna
GPT-5.6 Luna is OpenAI's fast, low-cost GPT-5.6 tier, generally available since 2026-07-09.
GPT-5.5 Cyber
⚠ Daybreak programme only - separate approval and provisioning. Apply at openai.com/daybreak. The earlier Daybreak cybersecurity model, still in OpenAI's 'Cyber models' pricing table at $12.50/M in…
GPT-5.5
OpenAI's flagship coding/reasoning model from April 2026, since followed by the GPT-5.6 family and GPT-6.
GPT-5.5 Pro
OpenAI's higher-accuracy Pro tier for harder questions.
GPT-5.4
Affordable GPT tier at half the GPT-5.5 standard price.
GPT-5.4 Pro
Pro tier for GPT-5.4.
GPT-5.4 mini
Smaller GPT-5.4 model for lower-latency coding, computer-use, and subagent workloads.
GPT-5.4 nano
Cheapest GPT-5.4-class model for classification, extraction, ranking, routing, and small subagents.
GPT-5.3-Codex
Specialized Codex model optimized for agentic coding tasks in Codex-like environments.
text-embedding-3-large LAUNCHED JAN 2024
Most capable OpenAI embedding model.
text-embedding-3-small LAUNCHED JAN 2024
Small OpenAI embedding model.
GPT-Rosalind
⚠ Trusted-access programme only - approved organisations doing internal life-sciences research. OpenAI's life-sciences reasoning model ('GPT-Rosalind: Life sciences reasoning for approved organizat…
chat-latest LAUNCHED MAY 2026
Alias for the latest Instant model used in ChatGPT.
GPT-4 LAUNCHED MAR 2023
Original GPT-4 (gpt-4-0613, formerly gpt-4-0314).
GPT-4 Turbo LAUNCHED NOV 2023
Latest snapshot was gpt-4-turbo-2024-04-09.
o4-mini LAUNCHED APR 2025
Successor to o3-mini — same headline rate but cheaper cache hit pricing ($0.275 vs $0.55).
GPT-3.5 Turbo LAUNCHED MAR 2023
Latest snapshot was gpt-3.5-turbo-0125 priced at $0.50/$1.50.
o3-mini LAUNCHED JAN 2025
First budget-tier reasoning model — 14x cheaper than original o1 ($1.10 vs $15 input) at competitive reasoning quality.
o1 LAUNCHED DEC 2024
First production o-series reasoning model (o1-2024-12-17 snapshot).
GPT-4.1 LAUNCHED APR 2025
OpenAI's GPT-4.1 series introduced 1M context windows to the GPT-4 family alongside another frontier price compression to $2/$8.
GPT-4.1 mini LAUNCHED APR 2025
Mid-tier GPT-4.1 — same 1M context as the base model, ~5x cheaper.
GPT-4o LAUNCHED MAY 2024
GPT-4o is the live alias rather than the frozen gpt-4o-2024-05-13 snapshot that OpenAI retires on 2026-10-23, and the two carry different prices - $2.50/$10 against $5/$15.
GPT-4o mini LAUNCHED JUL 2024
Set a new cheap-tier floor at OpenAI when it launched — $0.15/$0.60 was a 33x compression vs GPT-3.5 Turbo's original $5/$15-equivalent positioning and made OpenAI competitive with DeepSeek and Gem…
GPT-4 32K LAUNCHED MAR 2023
Extended-context GPT-4 variant.
GPT-4.5 preview LAUNCHED FEB 2025
Short-lived experimental flagship from OpenAI — only available for ~5 months.
GPT-4 Vision Preview LAUNCHED NOV 2023
First public OpenAI model with image input (alongside gpt-4-turbo's 128K context).
o1-mini LAUNCHED SEP 2024
Mini reasoning model launched alongside o1 — 5x cheaper than full o1 and surprisingly competitive on coding benchmarks.
| Model | Input /M | Output /M | Cached | Context | Max output | Vision | Tools | Tier |
|---|---|---|---|---|---|---|---|---|
| GPT-6.1 Sol | $2.00 | $10.00 | $0.10−95% | 1.05M | — | ✓ | ✓ | Frontier |
| GPT-6 Sol | $2.00 | $10.00 | $0.20−90% | 1.05M | — | ✓ | ✓ | Frontier |
| GPT-6 Luna | $0.10 | $0.50 | $0.01−90% | 1.05M | — | ✓ | ✓ | Light |
| GPT-6 Astra FLAGSHIP | $10.00 | $50.00 | $1.00−90% | 1.05M | — | ✓ | ✓ | Frontier |
| GPT-5.6 Cyber RESTRICTED | $12.50 | $75.00 | $1.25−90% | 400K | Daybreak programme only - separate approval and provisioning for approved defenders doing authorised vulnerability research. Apply at openai.com/daybreak. | ✓ | ✗ | Restricted |
| GPT-5.6 Sol | $4.00 | $20.00 | $0.40−90% | 1.05M | — | ✓ | ✓ | Frontier |
| GPT-5.6 Terra | $2.00 | $12.00 | $0.20−90% | 1.05M | — | ✓ | ✓ | Active |
| GPT-5.6 Luna | $0.20 | $1.20 | $0.02−90% | 1.05M | — | ✓ | ✓ | Light |
| GPT-5.5 Cyber RESTRICTED | $12.50 | $75.00 | $1.25−90% | not listed | Daybreak programme only - separate approval and provisioning. Apply at openai.com/daybreak. | ✗ | ✗ | Restricted |
| GPT-5.5 | $5.00 | $30.00 | $0.50−90% | 1M | — | ✓ | ✓ | Frontier |
| GPT-5.5 Pro | $30.00 | $180 | — | 1.05M | — | ✓ | ✓ | Active |
| GPT-5.4 | $2.50 | $15.00 | $0.25−90% | 1.05M | — | ✓ | ✓ | Mid |
| GPT-5.4 Pro | $30.00 | $180 | — | 1.05M | — | ✓ | ✓ | Active |
| GPT-5.4 mini | $0.75 | $4.50 | $0.075−90% | 400K | — | ✓ | ✓ | Light |
| GPT-5.4 nano | $0.20 | $1.25 | $0.02−90% | 400K | — | ✓ | ✓ | Active |
| GPT-5.3-Codex | $1.75 | $14.00 | $0.175−90% | 400K | — | ✓ | ✓ | Active |
| text-embedding-3-large | $0.13 | Free | — | input-only | — | ✗ | ✓ | Active |
| text-embedding-3-small | $0.02 | Free | — | input-only | — | ✗ | ✓ | Active |
| GPT-Rosalind RESTRICTED | $5.00 | $25.00 | $0.50−90% | not listed | Trusted-access programme only - approved organisations doing internal life-sciences research. | ✗ | ✗ | Restricted |
| chat-latest | $5.00 | $30.00 | $0.50−90% | 400K | — | ✓ | ✓ | Active |
| GPT-4 DEPRECATED | $30.00 | $60.00 | — | 8K | — | ✗ | ✓ | Deprecated |
| GPT-4 Turbo DEPRECATED | $10.00 | $30.00 | — | 128K | — | ✓ | ✓ | Deprecated |
| o4-mini DEPRECATED | $1.10 | $4.40 | $0.275−75% | 200K | — | ✗ | ✓ | Deprecated |
| GPT-3.5 Turbo DEPRECATED | $0.50 | $1.50 | — | 16K | — | ✗ | ✓ | Deprecated |
| o3-mini DEPRECATED | $1.10 | $4.40 | $0.55−50% | 200K | — | ✗ | ✓ | Deprecated |
| o1 DEPRECATED | $15.00 | $60.00 | $7.50−50% | 200K | — | ✗ | ✓ | Deprecated |
| GPT-4.1 LEGACY | $2.00 | $8.00 | $0.50−75% | 1M | — | ✓ | ✓ | Legacy |
| GPT-4.1 mini LEGACY | $0.40 | $1.60 | $0.10−75% | 1M | — | ✓ | ✓ | Legacy |
| GPT-4o LEGACY | $2.50 | $10.00 | $1.25−50% | 128K | — | ✓ | ✓ | Legacy |
| GPT-4o mini LEGACY | $0.15 | $0.60 | $0.075−50% | 128K | — | ✓ | ✓ | Legacy |
| GPT-4 32K RETIRED | — | — | — | Retired Jun 6, 2025 | → GPT-4o | ✗ | ✗ | Retired |
| GPT-4.5 preview RETIRED | — | — | — | Retired Jul 14, 2025 | → GPT-5.5 | ✓ | ✗ | Retired |
| GPT-4 Vision Preview RETIRED | — | — | — | Retired Dec 6, 2024 | → GPT-4o | ✓ | ✗ | Retired |
| o1-mini RETIRED | — | — | — | Retired Oct 27, 2025 | → GPT-5.6 Terra | ✗ | ✗ | Retired |
Pricing across the lineup.
The last 12 months at OpenAI.
Major releases, deprecations, and corporate events. Sourced from official OpenAI announcements, the developers.openai.com deprecation queue, and recent press.
$2/$10 card as GPT-6 Sol with the cache read halved to $0.10, and it takes 6 Sol's place in OpenAI's model list between Astra and Luna. The same day GPT-6 Astra gets an Ultrafast tier at six times standard, $60/$300$2/$10 against $4/$20, Luna $0.10/$0.50 against $0.20/$1.20. Long-context pricing still begins above 272K input tokens, where input doubles but output rises only by half$10 input, $1 cached and $50 output, exactly 2.5x GPT-5.6 Sol on every rate; rollout starts with Trusted Access Program enterprises, with API and paid ChatGPT plans said to follow within days$4 input, $0.40 cached and $20 output, marked promotional at least through November 21, 2026; the cheapest-to-frontier gap narrows from 25x to 17.9xplatform.openai.com — direct API, Playground, dashboard, billing, Batch + Flex + Priority tiers. Full pricing parity with everything you see on this page.
CLOUD · MICROSOFTModels in 28 Azure regions with EU + US Data Zones for in-region processing and storage. The default path for enterprises that need Microsoft contracts, BAAs, and Azure-native networking.
CONSUMER · WEBFree / Go / Plus / Pro / Business / Enterprise tiers at chatgpt.com plus iOS, Android, macOS, Windows. End-user product; no API access. Go from $8/mo, Plus $20/mo, Pro from $100/mo.
AGENT · CLIOpenAI's coding agent. Included on every ChatGPT plan including the free one; also runs on an API key, billed per token at the rates on this page.
OpenAI was founded on December 11, 2015 in San Francisco as a nonprofit research lab. The original founders included Sam Altman, Elon Musk, Ilya Sutskever, Greg Brockman, and Trevor Blackwell, with an initial $1B funding pledge from a group of donors. The stated mission: ensure that artificial general intelligence benefits all of humanity.
The corporate structure has gone through three eras. From 2015–2019 the lab ran as a pure 501(c)(3) nonprofit. In 2019 OpenAI created a capped-profit subsidiary (OpenAI LP, profit returns capped at 100×) to raise capital — Microsoft committed $1B that year and would eventually invest more than $13B for a 27% stake. In November 2023 the nonprofit board briefly removed and reinstated Sam Altman within five days, surfacing tension between the safety-focused board and the commercial entity. On October 28, 2025 the restructure completed: OpenAI converted into a public-benefit corporation with the OpenAI Foundation holding 26%, Microsoft holding 27%, and employees/investors holding 47%.
The technical contributions sit at the centre of the modern LLM era. The lab published the GPT-1 / GPT-2 / GPT-3 / GPT-4 / GPT-5 series, launched ChatGPT on November 30, 2022 (fastest product to 100M users), shipped DALL-E, Whisper, Sora, and the o-series reasoning models. As of September 2026 the API lineup centres on GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, with the GPT-5.6 family still on the price list and the older o-series and GPT-4 family queued for API retirement on Oct 23, 2026.
Funding scaled hard in 2025–2026. After a $157B valuation in October 2024 and a $300B round in early 2025, OpenAI closed a ~$122B Series F at a ~$852B valuation in April 2026, with Amazon committing $50B and SoftBank and Nvidia $30B each in February 2026. In parallel, the Stargate Project announced in January 2025 — a $500B AI infrastructure JV with SoftBank, Oracle, and MGX — has added 5 new US sites and ~7 GW of planned capacity through 2026.
Compared to Anthropic: larger, broader model surface (image gen, audio, video, embeddings), bigger consumer brand via ChatGPT, but weaker stated safety commitments (no published RSP-equivalent) and a longer history of leadership volatility. Compared to Google DeepMind: weaker integration with productivity suites and a smaller context window on the flagship.
Anthropic
Stronger on coding-agent benchmarks and zero-retention defaults. Smaller lineup, no native image gen, no consumer app at ChatGPT's scale.
CLOUD GIANT1M-token context across the Gemini line and native Workspace integration. Behind OpenAI on coding-agent benchmarks; ecosystem reach is the moat, not raw quality.
WILDCARDSpaceXAI
Memphis Colossus supercluster and X/Twitter integration. Smaller developer ecosystem and shorter pricing history; Grok is improving fast but starting from behind.
Meta
Llama 4 405B is the strongest open-weight option and is self-hostable on your own hardware. Trails frontier closed labs on coding and reasoning benchmarks by ~10pt.
DeepSeek
10–30× cheaper at frontier quality with off-peak discounts on top. China-based — geopolitical and data-residency considerations apply for some buyers.
OPEN-WEIGHT GIANTAlibaba (Qwen)
The Qwen (Tongyi Qianwen) model family from Alibaba Cloud's Tongyi Lab. First released in 2023, Qwen is the most-downloaded open-weight model family in the world — most tiers ship under Apache 2.0 on Hugging Face and ModelScope, while the proprietary Max tier is served through Alibaba Cloud's Model Studio.
Frequently asked.
Practical questions about OpenAI, the GPT lineup, and how to spend on it without surprises.
Q · 01 Which GPT model should I start with? +
$10.00/$50.00 per M) for complex reasoning and coding, choose GPT-6.1 Sol ($2.00/$10.00) to balance intelligence and cost, or GPT-6 Luna ($0.10/$0.50) for cost-sensitive, high-volume work. GPT-6 Sol and the GPT-5.6 family (Sol, Terra, Luna) stay on the price list.Q · 02 How safe is OpenAI relative to Anthropic? +
Q · 03 Does OpenAI offer EU data residency? +
Q · 04 What's the data retention policy? +
Q · 05 Has GPT-6 launched? +
Q · 06 Can I get a discount? +
Q · 07 How does OpenAI compare to other frontier labs? +
Q · 08 What are GPT-Rosalind and the Daybreak cyber models? +
$5.00/$25.00 per M, billed from October 5, 2026. GPT-5.6 Cyber and GPT-5.5 Cyber are for approved defenders in the Daybreak programme, at $12.50/$75.00. None of them is in this site's calculators.