Overview
ChatGPT is OpenAI's AI assistant for writing, research, coding, data analysis, documents, presentations, spreadsheets, and tool-using workflows. Its current flagship release is GPT-6 Astra, announced on September 3, 2026 for ChatGPT, the OpenAI API, and Amazon Bedrock.
Astra makes ChatGPT more capable on long-running agent tasks. For developers, the OpenAI API adds Astra-specific async tool calling, mid-turn steering, and cache-preserving reasoning updates; the model also improves computer use and supports a 1.05-million-token API context window. That makes it useful for complex coding and professional work, but it is neither the cheapest model nor an automatic upgrade for every workflow.
OpenAI is keeping the GPT-5.6 family relevant for routing. GPT-5.6 Sol, Terra, and Luna offer different capability and cost levels, while Astra becomes the premium path for tasks where better completion, long context, or agent control can justify higher API prices. Users should choose by task rather than treating the newest model as the default for all prompts.
Key Features
GPT-6 Astra for demanding work - Astra targets computer use, coding, science, security, and professional knowledge work. Eligible ChatGPT users get plan-dependent access, while developers can call
gpt-6-astrathrough the API.Long-running agent controls - In the OpenAI API, asynchronous tool calling lets Astra continue reasoning, call other tools, or handle independent work while the application executes a tool and manages pending work. Mid-turn steering uses a WebSocket Responses API connection, while cache-preserving reasoning updates can change effort between responses.
Broad tool support - The Responses API supports web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, and tool search with Astra. Production use still requires scoped permissions and reliable tool-result handling.
Large text and image context - Astra accepts text and image input, supports a 1,050,000-token context window, and can produce up to 128,000 output tokens. Audio and video inputs are not supported directly.
GPT-5.6 routing options - Sol remains a capable previous-generation flagship, while Terra and Luna support lower-cost and higher-volume work. This gives teams a practical model ladder instead of forcing all workloads onto Astra.
Work across documents and software - ChatGPT can support documents, presentations, spreadsheets, research, app building, debugging, and code review. Dedicated AI presentation tools may still fit teams that need template governance or deterministic slide production.
Pricing & Plans
ChatGPT has a free tier and multiple paid plans. In the U.S., Go is $8 per month, Plus is $20 per month, and Pro has $100 and $200 monthly tiers. Go, Plus, and Pro do not offer annual billing; Business supports monthly and annual billing, while Enterprise pricing is contact sales.
OpenAI's launch announcement described a broader staged rollout to Plus, Pro, Business, and Enterprise, but its current ChatGPT Help Center says GPT-6 Astra is not included with Plus in Chat. Current Chat access is instead rolling out as GPT-6 Pro for Pro $100, Pro $200, Business, and Enterprise, with Enterprise access also governed by workspace model-access permissions. Free, Go, and Plus do not currently include GPT-6 Pro in Chat.
Current included GPT-6 Pro Chat limits are 50 messages per week on Pro $100, 200 per week on Pro $200, 15 per month on Business Standard, and 50 per week on Business Premium. Managed-workspace availability and limits can additionally depend on workspace settings. ChatGPT Work and Codex use separate Astra allowances.
ChatGPT subscriptions and API usage are billed separately. Standard GPT-6 Astra API pricing is $10 per 1M input tokens, $1 per 1M cached-input reads, $12.50 per 1M cache-write tokens, and $50 per 1M output tokens. Batch and Flex are listed at 50% of Standard pricing, while Fast processing costs twice the applicable Standard rate.
Large-context use needs a separate budget check. When input exceeds 272,000 tokens, OpenAI applies twice the input and cache rates and 1.5 times the output rate to the full request. Teams should measure cost per completed task, including retries and review time, before routing repository-scale or document-heavy work to Astra by default.
GPT-5.6 remains useful where cost matters. OpenAI's current published standard rates are $4 input and $20 output per 1M tokens for Sol, with promotional pricing available at least through November 21, 2026; $2 input and $12 output for Terra; and $0.20 input and $1.20 output for Luna. Confirm live rates and availability before budgeting because promotional pricing and product routing can change.
How It Compares
Compared with GPT-5.6 Sol, Astra is a stronger choice for workflows that need computer use, long-context retrieval, concurrent tools, and interactive steering. OpenAI reports gains on OSWorld 2.0, Agents' Last Exam, Terminal-Bench, long-context retrieval, and cybersecurity evaluations. Those are vendor-published results, so production decisions should use private tasks and real tool schemas.
GPT-5.6 has the clearer cost advantage. At the current published promotional Sol rate, GPT-5.6 Sol is 60% cheaper than Astra on both input and output tokens, while Terra and Luna are cheaper again. Existing GPT-5.6 integrations also avoid Astra migration issues such as unsupported sampling parameters and Responses API requirements for tool calling.
Compared with standalone assistants, ChatGPT's advantage is breadth: chat, coding, research, files, images, artifacts, and API-backed agents sit within one ecosystem. The tradeoff is operational complexity across plan limits, model selection, API billing, regions, tools, and enterprise controls. For alternatives organized around switching reasons, see the ChatGPT alternatives guide.
Best For
- Individuals who want one assistant for writing, research, coding, files, and analysis.
- Engineering teams using ChatGPT and Codex for repository work, debugging, and review.
- Operations and knowledge-work teams producing documents, presentations, and data analysis.
- Developers building AI agents with the Responses API and multiple tools.
- Organizations that can route difficult tasks to Astra while keeping routine work on cheaper models.
Limitations & Safety
ChatGPT can produce incorrect facts, code, calculations, citations, or interface actions. Astra's April 30, 2026 knowledge cutoff means current information should use retrieval, and outputs with legal, medical, financial, security, or operational impact need qualified review.
OpenAI classifies Astra at its Critical cybersecurity capability threshold and applies stronger safeguards and monitoring. The company also reports lower written-reasoning monitorability than GPT-5.6 Sol. For production agents, rely on observable tool logs, sandboxing, least-privilege credentials, approval gates, and outcome checks rather than hidden reasoning as evidence of correctness.
FAQ
What is the current flagship model in ChatGPT?
GPT-6 Astra is OpenAI's current flagship release in this listing. It is rolling out to eligible ChatGPT paid plans and is also available through the OpenAI API and Amazon Bedrock where enabled.
Is GPT-6 Astra included with ChatGPT Plus?
No, not in Chat as of September 4, 2026. OpenAI's current Help Center says GPT-6 Astra is not included with ChatGPT Plus in Chat and is rolling out there as GPT-6 Pro for Pro $100, Pro $200, Business, and Enterprise. Enterprise access also depends on workspace model-access permissions, and rollout timing can vary by product and account.
Is ChatGPT free?
Yes, ChatGPT has a free tier. Advanced-model access, higher limits, work features, and GPT-6 Astra availability depend on the current plan and rollout.
How much does GPT-6 Astra cost in the API?
Standard rates are $10 per 1M input tokens, $1 for cached-input reads, $12.50 for cache writes, and $50 per 1M output tokens. Long-context and Fast processing can raise the applicable rates.
Should I use GPT-6 Astra instead of GPT-5.6?
Use Astra when better agent completion, computer use, coding, science, or long-context performance offsets its higher cost and migration effort. Keep GPT-5.6 for stable or cost-sensitive workflows until private evaluations show a clear benefit.
Does GPT-6 Astra support tool calling?
Yes, but OpenAI says Astra tool calling requires the Responses API. If enabling async tool calling, the application must manage pending work and return each tool result using the original call_id; add any cancellation or concurrency handling required by the application's own orchestration design.
