OpenAI introduces GPT-6.2 Astra as its frontier model for the hardest projects, released on November 17, 2026. It accepts text and image inputs across a 1,050,000-token context window (up to 922,000 input tokens and 128,000 output tokens) and delivers cleaner code, better frontends and 3D models, and better results on demanding tasks with fewer output tokens. On Kie.ai it currently supports chat requests only.

Pricing: Input 480 credits / 1M tokens (≈ $2.40), Cached Input 24 credits / 1M tokens (≈ $0.12), Cache Writes 600 credits / 1M tokens (≈ $3.00), Output 2400 credits / 1M tokens (≈ $12.00) — all 70% below official pricing. High-tier top-ups come with bonus credits that lower the effective price by about 10%.
gpt-6-2-astra

Automatically append previous messages to maintain multi-turn context. May increase token usage.

Function calling
Web Access
Reasoning effort
Stream Response

How can I help you today?

README

Complete guide to using gpt-6-2-astra

GPT-6.2 Astra API for Complex Engineering, Science and Professional Work

GPT-6.2 Astra accepts text and images and responds in text across a 1,050,000-token context window, with up to 922,000 input tokens and 128,000 output tokens. It writes cleaner code, builds more capable web frontends and 3D designs, and produces professional deliverables with fewer output tokens.
curl https://api.kie.ai/codex/v1/responses \
-H "Authorization: Bearer $KIE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-2-astra",
"input": "Design an idempotent webhook handler for payment retries.",
"reasoning": { "effort": "medium" },
"stream": false
}'
Reasoning

Key Capabilities of GPT-6.2 Astra

OpenAI positions GPT-6.2 Astra as its flagship model for demanding engineering, scientific research and multistep professional workflows.
Professional Slides, Workbooks and Documents
01

Professional Slides, Workbooks and Documents

OpenAI trained GPT-6.2 Astra to produce clearer slides, spreadsheets and documents that follow the conventions of law, sales, finance and accounting with less filler. Given an original slide file or a screenshot, it matches layout, colors and typography down to the exact font size. On Kie.ai the model returns text (for example structured content or code), not slide or spreadsheet files.
Cleaner Code, Web Frontends and 3D Design
02

Cleaner Code, Web Frontends and 3D Design

GPT-6.2 Astra implements and tests complex changes in existing codebases with cleaner code style and more focused tests. It brings stronger visual reasoning to web design, writes code for individual parts and connections in CAD workflows, and builds interactive 3D browser experiences; on AA-3D, GPT-6.2 Astra ranks first with 1,318 Elo.
1,050,000-Token Context and Up to 128,000 Output Tokens
03

1,050,000-Token Context and Up to 128,000 Output Tokens

Each request supports a 1,050,000-token context window with up to 922,000 input tokens and up to 128,000 output tokens. The model accepts text and image inputs and responds with text, with a May 18, 2026 knowledge cutoff.
Five Reasoning Effort Levels From Low to Max
04

Five Reasoning Effort Levels From Low to Max

On Kie.ai's /codex/v1/responses endpoint, guide how much the model thinks with reasoning.effort set to low, medium, high, xhigh or max.
Scientific Research and Defensive Security
05

Scientific Research and Defensive Security

GPT-6.2 Astra sets new highs on Terminal-Bench-Science 0.1 (+9 percentage points over GPT-6 Astra) and the Vals Science Index for interpreting biological data and running experiments on simulated systems. OpenAI also trained it to write secure code from the start and produce patches that preserve intended behavior.
Responses API Tools and Preserved Reasoning
06

Responses API Tools and Preserved Reasoning

On Kie.ai, /codex/v1/responses supports custom function calling or built-in web_search (choose one per request), plus text, image and file inputs. OpenAI recommends keeping prompt prefixes unchanged and appending new instructions to help preserve reasoning across turns and keep earlier input eligible for prompt caching.
SpecificationGPT-6.2 Astra
DeveloperOpenAI
Model IDgpt-6-2-astra on Kie.ai (gpt-6.2-astra also accepted)
Release dateNovember 17, 2026
ModalitiesText and image input; text output (audio and video unsupported)
Context window1,050,000 tokens (up to 922,000 input tokens)
Max output tokens128,000 tokens
Reasoning effortlow, medium, high, xhigh, max
Kie.ai endpoint & toolsPOST /codex/v1/responses; custom function calling or web_search (one per request); text, input_image and input_file
Kie.ai featuresStreaming (default), function calling or web search (one per request), image and file input, prompt caching
Unsupported parameters and featuresParameters: temperature, top_p, logprobs, top_logprobs; features: fine-tuning, predicted outputs
Knowledge cutoffMay 18, 2026
BenchmarkGPT-6.2 Astra ResultWhat It Measures
Agents' Last Exam+5.2 points over GPT-6 Astra (new high)Long-horizon multistep professional workflows
Terminal-Bench-Science 0.1+9 percentage points over GPT-6 Astra (new high)Scientific work carried out in a terminal
GDPval v2+6.2 percentage points over GPT-6 Astra with 22% fewer output tokensMultistep consulting, finance and design assignments
AA-AnalystAgent62.5% of questions answered correctly on all 5 independent attempts (new high)Financial and expenditure analysis over real spreadsheets and reports
AA-3D1,318 Elo, ranked first (GPT-6.2 Sol: 1,259 Elo)Blinded human preference on interactive 3D browser builds
BenchCAD 2.1 Core32% improvement over GPT-6 AstraBuilding 3D parts and assemblies from engineering drawings
Cybersecurity evaluationsSystem Card targets: 36.4% solved vs 18.2% for GPT-6 Astra; ExploitBench Internal Port: +22 points over GPT-6 AstraSeparate security evaluations reported by OpenAI

How to Work With GPT-6.2 Astra

Four OpenAI API practices that govern accuracy, tool use and token efficiency.
  1. 1

    Step 1: Choose Between GPT-6.2 Astra and GPT-6.2 Sol

    Use GPT-6.2 Astra for your hardest projects across engineering, scientific research, professional deliverables, frontends and 3D design. For day-to-day coding, OpenAI recommends starting with GPT-6.2 Sol, its efficient, versatile model for everyday work.
  2. 2

    Step 2: Set Reasoning Effort From Low to Max

    On Kie.ai, pass reasoning.effort as low, medium, high, xhigh or max in /codex/v1/responses. OpenAI recommends starting with medium for a new integration, or switching legacy none and minimal settings to low. Omit temperature, top_p, logprobs and top_logprobs.
  3. 3

    Step 3: Keep Prompt Prefixes Unchanged

    Starting with GPT-6.2, OpenAI's API drops reasoning items when it detects mid-conversation prefix edits (such as changes to instructions or tool calls), a model downgrade, or a change in the end user or organization. OpenAI recommends keeping prompt prefixes unchanged and appending new instructions to help preserve reasoning across turns and keep earlier input eligible for prompt caching.
  4. 4

    Step 4: Call /codex/v1/responses for Tools and Multimodal Inputs

    On Kie.ai, send requests to POST /codex/v1/responses with custom function definitions or {"type":"web_search"} (choose one tool type per request). Because GPT-6.2 Astra writes more concisely and directly by default, review legacy system prompts that repeat "be concise" instructions.

What You Can Build With GPT-6.2 Astra

Typical workloads for OpenAI's frontier model with a 1.05M-token context window.

Complex Codebase Changes and Security Patches

Implement and test multi-file changes across large repositories, write secure code from the start and generate patches that preserve intended behavior.

Financial Workbooks, Legal Briefs and Slide Decks

Analyze figures across spreadsheets and expenditure reports, find facts in legal source documents and draft structured slide content or code that matches your existing templates.

Interactive Web Frontends and 3D Browser Apps

Turn written briefs and visual references into distinctive web interfaces, browser games, code for CAD mechanical models and interactive 3D scenes.

Scientific Computing and Data Interpretation

Write Python solvers for physical simulations, interpret biological or experimental datasets, and connect your own execution tools via function calling for terminal-based scientific workflows.

GPT-6.2 Astra on Kie.ai

01

Quick Start on Kie.ai

Try GPT-6.2 Astra in the playground on this page, or send a POST request to https://api.kie.ai/codex/v1/responses with Authorization: Bearer YOUR_API_KEY and a JSON body such as {"model":"gpt-6-2-astra","input":"Summarize the key assumptions","reasoning":{"effort":"medium"},"stream":false}.
02

Chat Models in the Kie.ai Market

Browse the chat models in the Kie.ai market, including GPT-6 Astra, GPT-6.1 Sol and GPT-6.1 Terra, alongside the GPT-6.2 Sol overview.
03

One Key Across the Kie.ai Market

All models in the Kie.ai market use the same authentication method: an Authorization: Bearer YOUR_API_KEY header with the key from your Kie.ai API Key Management page.
04

One Credit Balance

Your account has one credit balance that all models in the Kie.ai market draw from, and you can check it with the Get Remaining Credits API.

Frequently Asked Questions

What is GPT-6.2 Astra?

GPT-6.2 Astra is OpenAI's frontier model for demanding engineering, scientific research and complex professional deliverables, released on November 17, 2026 alongside GPT-6.2 Sol. On Kie.ai it is available as gpt-6-2-astra (or gpt-6.2-astra) and currently supports chat requests only.

How is GPT-6.2 Astra different from GPT-6.2 Sol?

Both models share a 1,050,000-token context window and a 128,000-token output limit, with text and image inputs and text output. OpenAI positions GPT-6.2 Astra as its flagship model for your hardest projects, and GPT-6.2 Sol as its efficient, versatile go-to model for day-to-day coding and everyday work.

What are the context window and token limits of GPT-6.2 Astra?

GPT-6.2 Astra has a 1,050,000-token context window, accepts up to 922,000 input tokens and generates up to 128,000 output tokens. Its knowledge cutoff is May 18, 2026.

Which reasoning effort levels does GPT-6.2 Astra support?

On Kie.ai, set reasoning.effort to low, medium, high, xhigh or max in requests to /codex/v1/responses. OpenAI recommends medium for new integrations and low when migrating from none or minimal.

Which endpoint and tools does GPT-6.2 Astra use on Kie.ai?

On Kie.ai, GPT-6.2 Astra runs on the OpenAI Responses-compatible endpoint POST /codex/v1/responses and currently supports chat requests only. You can pass custom function tools or {"type":"web_search"} (choose one per request), along with text, image URLs and file URLs.

Why does OpenAI recommend keeping prompt prefixes unchanged?

Starting with GPT-6.2, OpenAI's API drops reasoning items when it detects mid-conversation prefix edits (such as changes to instructions or tool calls), a model downgrade, or a change in the end user or organization. OpenAI recommends keeping prompt prefixes unchanged and appending new instructions to help preserve reasoning across turns and keep earlier input eligible for prompt caching.

What inputs and sampling parameters are unsupported in GPT-6.2 Astra?

GPT-6.2 Astra accepts text and image inputs and outputs text; audio and video modalities, fine-tuning and predicted outputs are unsupported. Omit temperature, top_p, logprobs and top_logprobs when sending requests.

How do I call the GPT-6.2 Astra API on Kie.ai?

Create an API key on your Kie.ai API Key Management page and send a POST request to https://api.kie.ai/codex/v1/responses with Authorization: Bearer YOUR_API_KEY and model set to gpt-6-2-astra (or gpt-6.2-astra). Responses stream as Server-Sent Events by default; set stream to false for a single JSON response, and choose either web_search or function tools per request. Full request schemas are in the GPT-6.2 Astra API docs.

How is the GPT-6.2 Astra API billed on Kie.ai?

Usage on Kie.ai is billed on a pay-as-you-go basis per million tokens across input, cached input, cache writes and output, with no subscription and credits that never expire. Current rates are shown in the pricing card on this page.