OpenAI's GPT-6 generation is sold as four named tiers, Astra, Sol, Luna and Terra, whose list prices differ by up to 100 times while their context windows are identical, so picking the wrong tier changes cost and behaviour far more than capacity. This page compares the five GPT-6 API models OpenAI documents: GPT-6 Astra, GPT-6 Sol, GPT-6.1 Sol, GPT-6 Luna and GPT-6.1 Terra. OpenAI's pages list no model called GPT-6 Terra; the Terra tier in this generation is GPT-6.1 Terra, and Sol exists as both GPT-6 Sol and its newer GPT-6.1 Sol. Every row uses identical fields: API ID, release date, context, cutoff, prices by processing tier, reasoning effort, tools, residency, rate limits, system card classifications and lifecycle. It is written for developers and buyers choosing a tier and for architects routing work between tiers. Models are listed by release date, newest first, an objective order that implies no ranking.
Last checked: 2026-11-15.
This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. It uses OpenAI's public API documentation as of November 2026. Model details change, so confirm against OpenAI's current documentation.
Key takeaways
- Five API models make up four tiers. Astra (3 September 2026), GPT-6 Sol and Luna (22 September), GPT-6.1 Sol (29 September) and GPT-6.1 Terra (26 October); GPT-6 Sol's page points to GPT-6.1 Sol as the newer Sol model.
- Context is the same across tiers. Every model lists 1,050,000 tokens of context and 128,000 tokens of output, with knowledge cutoffs between 20 April and 1 June 2026.
- Standard prices run from $0.10 to $50. Per million tokens: Luna $0.10/$0.50, GPT-6.1 Terra $1/$5, both Sol models $2/$10, Astra $10/$50.
- API behaviour differs by tier. Astra and GPT-6.1 Sol reject reasoning effort none; GPT-6 Sol and Luna allow Chat Completions function calling only at effort none; Ultrafast is priced only for Astra and GPT-6.1 Sol.
- Three models are classed Critical in cyber. OpenAI's system cards treat Astra, GPT-6.1 Sol and GPT-6.1 Terra as Critical in cybersecurity, and the October GPT-6 Sol and Luna as High, below Critical.
What is the GPT-6 family?
GPT-6 is OpenAI's model generation that followed the GPT-5.6 family (Sol, Terra and Luna, released 9 July 2026). OpenAI's migration guide describes gpt-6-astra for "most demanding work", gpt-6.1-sol for "complex work at lower cost", gpt-6.1-terra for "visual reasoning and writing" and gpt-6-luna for "focused, high-volume tasks". The tier names carry over from GPT-5.6, which had Sol, Terra and Luna; Astra is new in GPT-6. On 7 October 2026 OpenAI's chat-latest update recommended "the GPT-6 model family" for production use.
All five models are proprietary, with no published weights or parameter counts. Each has a single undated snapshot on its model page, so the ID can point at updated weights: OpenAI's GPT-6 October system card describes October versions of GPT-6 Sol and Luna that replaced GPT-5.6 Sol and Luna in ChatGPT, while the September versions remained in Codex and ChatGPT Work. For the move from the previous flagship, see GPT-6 vs GPT-5.5; for the whole line-up, the OpenAI Models hub.
Every model at a glance
As of 2026-11-15. Source: OpenAI model pages and API changelog.
| Model | API ID | Released | Status on OpenAI pages |
|---|---|---|---|
| GPT-6.1 Terra | gpt-6.1-terra | 2026-10-26 | Current; no deprecation entry |
| GPT-6.1 Sol | gpt-6.1-sol | 2026-09-29 | Current; no deprecation entry |
| GPT-6 Luna | gpt-6-luna | 2026-09-22 | Current; no deprecation entry |
| GPT-6 Sol | gpt-6-sol | 2026-09-22 | Page points to GPT-6.1 Sol as the newer Sol model; not on the pricing page; no deprecation entry |
| GPT-6 Astra | gpt-6-astra | 2026-09-03 | Current; no deprecation entry |
All five take text and image input and return text; audio and video are not supported on any model page. Fine-tuning is not supported on any of them. Parameters and licence terms beyond "proprietary" are not stated by OpenAI.
When was each model released?
As of 2026-11-15. Dates as shown in OpenAI's API changelog.
| Date | Event | API ID |
|---|---|---|
| 2026-09-03 | GPT-6 Astra released; async tool calling, mid-turn steering and mid-conversation reasoning effort changes added in the Responses API | gpt-6-astra |
| 2026-09-22 | GPT-6 Sol and GPT-6 Luna released as reasoning models with text and image input | gpt-6-sol, gpt-6-luna |
| 2026-09-25 | Image encoding bug that degraded image understanding in GPT-6 Sol and Luna fixed | gpt-6-sol, gpt-6-luna |
| 2026-09-29 | GPT-6.1 Sol released with Multi-agent beta; Ultrafast mode added for Astra with US data residency only | gpt-6.1-sol, gpt-6-astra |
| 2026-10-07 | October versions of GPT-6 Sol and Luna documented in a new system card; chat-latest recommends the GPT-6 family | gpt-6-sol, gpt-6-luna |
| 2026-10-08 | Ultrafast mode added for GPT-6.1 Sol in the Responses API | gpt-6.1-sol |
| 2026-10-16 | Ultrafast mode added to the Agents API via agent.service_tier | Supported models |
| 2026-10-26 | GPT-6.1 Terra released | gpt-6.1-terra |
How much does each model cost?
As of 2026-11-15. USD per million text tokens from OpenAI's pricing page and model pages. The pricing page's tabs are labelled Standard, Flex/Batch, Fast and Ultrafast, but the later tables are headed only "All models"; columns follow the tab order and match each model page's multipliers (Batch and Flex 50%, Fast 2x, Ultrafast 6x where offered).
| Model | Standard input / cached / output | Cache write 30m / 12h | Flex or Batch input / output | Fast input / output | Ultrafast input / output |
|---|---|---|---|---|---|
| GPT-6.1 Terra | $1.00 / $0.05 / $5.00 | $1.25 / $2.50 | $0.50 / $2.50 | $2.00 / $10.00 | Not priced on the pricing page |
| GPT-6.1 Sol | $2.00 / $0.10 / $10.00 | $2.50 / $5.00 | $1.00 / $5.00 | $4.00 / $20.00 | $12.00 / $60.00 |
| GPT-6 Luna | $0.10 / $0.01 / $0.50 | $0.125 / $0.25 | $0.05 / $0.25 | $0.20 / $1.00 | Not priced on the pricing page |
| GPT-6 Sol | $2.00 / $0.20 / $10.00 (model page) | $2.50 / not stated | 50% of Standard (model page) | 2x Standard (model page) | Not stated on the model page |
| GPT-6 Astra | $10.00 / $1.00 / $50.00 | $12.50 / $25.00 | $5.00 / $25.00 | $20.00 / $100.00 | $60.00 / $300.00 |
Common rules: each model page bills prompts over 272K input tokens at 2x input and cache rates and 1.5x output for the full request. Regional processing (data residency) endpoints carry a 10% uplift, as do FedRAMP endpoints according to the pricing page. GPT-6.1 Sol's page states subscription usage in Codex and ChatGPT Work has no long-context surcharge. Cached input is 5% of the input rate on GPT-6.1 Sol and GPT-6.1 Terra and 10% on Astra, Luna and GPT-6 Sol. Tool calls such as web search carry separate per-call fees on the pricing page.
Worked cost example
At Standard prices, 10,000 requests a day with a 20K-token prompt and a 1K-token response use 200 million input and 10 million output tokens. Daily cost: Luna $20 plus $5, $25; GPT-6.1 Terra $200 plus $50, $250; either Sol model $400 plus $100, $500; Astra $2,000 plus $500, $2,500. If 15K tokens of each prompt are a cached prefix, GPT-6.1 Sol's input falls to $15 cached plus $100 uncached, while GPT-6 Sol's falls to $30 plus $100, before cache write charges. Re-run with your own token counts. For cross-provider prices, see the LLM Pricing Tracker.
What changed between models?
Only changes documented on OpenAI's model pages, changelog and migration guide are listed.
- GPT-6 Sol to GPT-6.1 Sol: cached input fell from $0.20 to $0.10; reasoning effort none was removed (the migration guide says to use low); the knowledge cutoff moved from 20 April to 30 April 2026; Multi-agent beta, reasoning.context all_turns and Ultrafast mode were added; Fast mode gained EU data residency support.
- GPT-5.6 Terra to GPT-6.1 Terra: the migration guide advises comparing the two on representative images and workflows at the same reasoning effort. The GPT-6.1 Terra system card reports HealthBench Professional of 62.7 against 57.7 for GPT-5.6 Terra.
- Astra as a new tier: released first, without reasoning effort none and with custom temperature, top_p and logprobs removed; the migration guide bases its prompting guidance on observed gpt-6-astra behaviour.
- Luna: keeps reasoning effort none and Chat Completions function calling at that setting, and has the highest token-per-minute limits in the family.
The migration guide documents these for the GPT-6 family as a whole: removing temperature, top_p and top_logprobs when reasoning effort is not none; async tool calling; and the configuration_update input item that changes effort mid-conversation without rewriting the prompt prefix. See Reasoning Models Explained.
What workloads does OpenAI position each model for?
This section records the provider's own positioning wording; it is not an assessment.
| Model | Positioning stated by provider | Related notes stated by provider |
|---|---|---|
| GPT-6.1 Terra | "A fast, efficient model for visual reasoning, precise computer use, and clear, natural writing" | Examples: matching product photos to catalogue entries, interpreting technical drawings, acting on application screens, drafting or revising text |
| GPT-6.1 Sol | "Near-Astra performance for complex work at a lower cost" | Complex coding, computer use and professional work |
| GPT-6 Luna | "Our most efficient model for focused, high-volume tasks" | Reasoning rated high, speed fast, price low on the model page |
| GPT-6 Sol | "Built to power complex coding and agentic workflows" | Newer Sol model is GPT-6.1 Sol |
| GPT-6 Astra | "Our most capable model for the most demanding work" | Complex reasoning, coding, computer use, research and document creation |
What is the context window of each model?
As of 2026-11-15. From each OpenAI model page.
| Model | Context window | Max output | Knowledge cutoff | Input | Output | Reasoning controls |
|---|---|---|---|---|---|---|
| GPT-6.1 Terra | 1,050,000 tokens | 128,000 tokens | 2026-06-01 | Text, image | Text | none, low, medium (default), high, xhigh, max |
| GPT-6.1 Sol | 1,050,000 tokens (500K default in Codex and ChatGPT Work) | 128,000 tokens | 2026-04-30 | Text, image | Text | low, medium (default), high, xhigh, max |
| GPT-6 Luna | 1,050,000 tokens | 128,000 tokens | 2026-05-18 | Text, image | Text | none, low, medium (default), high, xhigh, max |
| GPT-6 Sol | 1,050,000 tokens | 128,000 tokens | 2026-04-20 | Text, image | Text | none, low, medium (default), high, xhigh, max |
| GPT-6 Astra | 1,050,000 tokens | 128,000 tokens | 2026-04-30 | Text, image | Text | low, medium, high, xhigh, max; default not stated on the model page |
Because the window is the same, tier choice does not change how much a request can hold; it changes price, effort options and the 272K long-context surcharge each request pays.
Capabilities and tools compared
As of 2026-11-15. All five model pages list the same ten Responses API tools: web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search. Differences are in how tools are called, residency and cloud IDs.
| Model | Function calling and structured outputs | Computer use tool | Fine-tuning | Cloud IDs stated by provider |
|---|---|---|---|---|
| GPT-6.1 Terra | Both supported; tool calling requires the Responses API | Supported | Not supported | Amazon Bedrock openai.gpt-6.1-terra; Google Cloud gpt-6.1-terra (private preview) |
| GPT-6.1 Sol | Both supported; tool calling requires the Responses API | Supported | Not supported | Amazon Bedrock openai.gpt-6.1-sol; Google Cloud gpt-6.1-sol (private preview) |
| GPT-6 Luna | Both supported; Chat Completions function calling only at effort none | Supported | Not supported | Amazon Bedrock openai.gpt-6-luna; Google Cloud gpt-6-luna (private preview) |
| GPT-6 Sol | Both supported; Chat Completions function calling only at effort none | Supported | Not supported | Amazon Bedrock openai.gpt-6-sol |
| GPT-6 Astra | Both supported; tool calling requires the Responses API | Supported | Not supported | Amazon Bedrock openai.gpt-6-astra; Google Cloud gpt-6-astra (private preview) |
Processing modes and data residency
| Model | Fast mode residency | Ultrafast mode | Model-specific agent features stated by provider |
|---|---|---|---|
| GPT-6.1 Terra | EU residency supported with Fast | Not priced on the pricing page | None beyond the family features in the migration guide |
| GPT-6.1 Sol | EU residency supported with Fast | US and EU residency plus global processing | Multi-agent beta; reasoning.context all_turns |
| GPT-6 Luna | EU residency for Standard, Fast, Flex and Batch | Not priced on the pricing page | None beyond the family features in the migration guide |
| GPT-6 Sol | EU residency for Standard, Flex and Batch; not with Fast | Not stated on the model page | None beyond the family features in the migration guide |
| GPT-6 Astra | EU residency not supported with Fast; Fast has no latency SLA | US residency and global processing only | Async tool calling, mid-turn steering and mid-conversation effort changes added at launch (changelog) |
For the security side of these tools, see Computer-Use and Browser Agents and Securing OpenAI Agents SDK and Responses API Guide.
Where is each model available?
As of 2026-11-15. Rate limits are the Build and Grow tiers from each model page; the Free tier is not supported for any GPT-6 model.
| Model | Availability stated by provider | Build tier (RPM / TPM) | Grow tier (RPM / TPM) |
|---|---|---|---|
| GPT-6.1 Terra | OpenAI API; Amazon Bedrock; Google Cloud private preview | 5,000 / 1,000,000 | 15,000 / 40,000,000 |
| GPT-6.1 Sol | OpenAI API; Codex and ChatGPT Work (500K default context); Amazon Bedrock; Google Cloud private preview | 5,000 / 1,000,000 | 15,000 / 40,000,000 |
| GPT-6 Luna | OpenAI API; ChatGPT (October version, per system card); Amazon Bedrock; Google Cloud private preview | 5,000 / 2,000,000 | 30,000 / 180,000,000 |
| GPT-6 Sol | OpenAI API; ChatGPT (October version), Codex and ChatGPT Work (September version), per system card; Amazon Bedrock | 5,000 / 1,000,000 | 15,000 / 40,000,000 |
| GPT-6 Astra | OpenAI API; Amazon Bedrock; Google Cloud private preview | 5,000 / 1,000,000 | 15,000 / 40,000,000 |
ChatGPT plan-level availability per model is not stated on the API pages checked. Partner clouds publish their own catalogues and dates.
Model cards and system cards
As of 2026-11-15. Classifications as stated in each system card under OpenAI's Preparedness Framework.
| Model | System card (published) | Cybersecurity | Biological and chemical | AI self-improvement |
|---|---|---|---|---|
| GPT-6.1 Terra | GPT-6.1 Terra (2026-10-26) | Critical | High | Below High |
| GPT-6.1 Sol | GPT-6.1 Sol (2026-09-29) | Critical | High | Below High |
| GPT-6 Luna (October) | GPT-6 October (2026-10-07) | High, below Critical | High | Below High |
| GPT-6 Sol (October) | GPT-6 October (2026-10-07) | High, below Critical | High | Below High |
| GPT-6 Astra | GPT-6 Astra (2026-09-03, change log to 2026-10-30) | Critical; "our first model to reach the Critical level" | High | Below High |
The September versions of GPT-6 Sol and Luna were covered in an appendix added to the Astra card on 2026-09-22. The October card states Luna did not need separate Critical testing because it scored below GPT-5.6 Sol on all High capability evaluations. See the Frontier AI Model Risk Frameworks Guide.
Benchmarks
As of 2026-11-15. Provider-reported figures from OpenAI's system cards. HealthBench (length-adjusted) is reported for every model; SEC-Bench Pro is reported for four of the five, in two different cards with different formats (pass@1 percentages to one or two decimal places). Card dates differ, so treat cross-row comparisons with care. Independent leaderboards (LMArena, Artificial Analysis, Epoch AI) were not checked for this page.
| Model | HealthBench Professional | HealthBench | HealthBench Hard | SEC-Bench Pro | Source and date |
|---|---|---|---|---|---|
| GPT-6.1 Terra | 62.7 | 55.1 | 31.4 | Not stated by OpenAI on the pages checked (2026-11-15) | GPT-6.1 Terra card, 2026-10-26 |
| GPT-6.1 Sol | 64.2 | 58.5 | 36.2 | 78.8% | GPT-6.1 Sol card, 2026-09-29 |
| GPT-6 Luna (October) | 48.2 | 50.4 | 27.2 | 48.50% | GPT-6 October card, 2026-10-07 |
| GPT-6 Sol (October) | 55.2 | 52.1 | 27.8 | 68.85% | GPT-6 October card, 2026-10-07 |
| GPT-6 Astra | 64.7 | 58.3 | 36.6 | 85.4% (reported in the GPT-6.1 Sol card) | GPT-6 Astra card, 2026-09-03 (HealthBench revised 2026-09-22) |
The GPT-6.1 Sol card also reports 66.3% on SEC-Bench Pro for GPT-6 Sol, a different figure from the October card's 68.85%, which applies to the October version.
Deprecations and retirements
As of 2026-11-15. Source: OpenAI deprecations page. No GPT-6 model has a deprecation entry. The page names GPT-6 models as replacements for retiring models, which shows which tier OpenAI maps each older model to.
| Model | API ID | Notice or deprecation date | Retirement or shutdown date | Stated replacement |
|---|---|---|---|---|
| GPT-5, o3 | gpt-5, o3 | 2026-06-11 | 2026-12-11 | gpt-6.1-sol |
| GPT-5 Pro, o3-pro | gpt-5-pro, o3-pro | 2026-06-11 | 2026-12-11 | gpt-6.1-sol with reasoning.mode: pro |
| GPT-5 mini | gpt-5-mini | 2026-06-11 | 2026-12-11 | gpt-6.1-terra |
| GPT-5 nano | gpt-5-nano | 2026-06-11 | 2026-12-11 | gpt-6-luna |
| GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini | Dated snapshots | 2026-10-14 | 2027-01-14 | gpt-6-luna |
| GPT-5.1, GPT-5.3 Codex | gpt-5.1, gpt-5.3-codex | 2026-10-01 | 2027-04-01 | gpt-6-sol |
| GPT-5.4 nano | gpt-5.4-nano | 2026-10-01 | 2027-04-01 | gpt-6-luna |
No retiring model is mapped to gpt-6-astra. See the LLM Deprecation and Retirement Tracker, the OpenAI Mini and Nano Models Compared page and the AI Model Retirement Guide.
Data handling and security
OpenAI's data controls page applies to every tier: API data is not used for training unless the customer opts in, abuse monitoring logs are kept for up to 30 days by default, Zero Data Retention and Modified Abuse Monitoring need OpenAI's approval, and OpenAI may make specific models ineligible for those arrangements for particular customers with advance written notice. Residency limits differ by tier and processing mode (see the processing modes table): Astra's Ultrafast mode supports US residency only. OpenAI's changelog lists misalignment monitoring among Astra's migration notes.
Routing work across tiers usually means one integration calling several model IDs with the same keys. Inventory which tiers each key and project can call, and set spend limits per project, since one leaked key can run Astra at Ultrafast prices. See the OpenAI API Keys Security Guide, How to set spend limits on LLM API keys and the LLMjacking Guide.
How to choose a model: factual criteria
| Criterion | What the documentation shows |
|---|---|
| Standard price per million tokens | Luna $0.10/$0.50; GPT-6.1 Terra $1/$5; GPT-6 Sol and GPT-6.1 Sol $2/$10; Astra $10/$50 |
| Need reasoning effort none | GPT-6.1 Terra, Luna and GPT-6 Sol |
| Function calling in Chat Completions | GPT-6 Sol and Luna, at effort none only |
| Ultrafast processing | Priced for Astra and GPT-6.1 Sol |
| EU residency with Fast mode | GPT-6.1 Sol, GPT-6.1 Terra and Luna |
| Multi-agent beta | GPT-6.1 Sol |
| Highest throughput limits | Luna: 30,000 RPM and 180M TPM at Grow |
| Latest knowledge cutoff | GPT-6.1 Terra, 2026-06-01 |
| Provider-stated focus | Astra demanding work; Sol complex coding and work; Terra visual reasoning, computer use and writing; Luna high-volume tasks |
Test candidate tiers on your own prompts at matched effort settings; system card evaluations differ by card and date.
Practitioner checklist
- Pin each workload to an explicit GPT-6 model ID and record it, with its tier and price, in the AI register.
- Move any gpt-6-sol workloads that send effort none to low before switching to gpt-6.1-sol, and retest.
- Use the Responses API for tool calling on Astra, GPT-6.1 Sol and GPT-6.1 Terra.
- Check data residency against the processing mode you plan to use, especially Fast and Ultrafast on Astra and GPT-6 Sol.
- Model cost with the 272K long-context multiplier, 30-minute and 12-hour cache writes and the 10% regional uplift.
- Map retiring models to OpenAI's stated GPT-6 replacements and finish migrations before the 11 December 2026 shutdowns.
- Set per-project spend limits and restrict which model tiers each key can reach.
Standards and references
- OpenAI: Models (accessed 2026-11-15)
- OpenAI: GPT-6 Astra model page (accessed 2026-11-15)
- OpenAI: GPT-6.1 Sol model page (accessed 2026-11-15)
- OpenAI: GPT-6 Sol model page (accessed 2026-11-15)
- OpenAI: GPT-6 Luna model page (accessed 2026-11-15)
- OpenAI: GPT-6.1 Terra model page (accessed 2026-11-15)
- OpenAI: Using the latest model (accessed 2026-11-15)
- OpenAI: Pricing (accessed 2026-11-15)
- OpenAI: API changelog (accessed 2026-11-15)
- OpenAI: Deprecations (accessed 2026-11-15)
- OpenAI: Data controls in the OpenAI platform (accessed 2026-11-15)
- OpenAI: GPT-6 Astra system card (accessed 2026-11-15)
- OpenAI: GPT-6.1 Sol system card (accessed 2026-11-15)
- OpenAI: GPT-6 October system card (accessed 2026-11-15)
- OpenAI: GPT-6.1 Terra system card (accessed 2026-11-15)
Related NHI Mgmt Group resources: OpenAI Models, GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, GPT-6.1 Terra, Frontier AI Models Compared.