OpenRouter 模型 ↗ · 2026-09-05
openai/gpt-6-astra
Model release OpenRouter 模型 Source language: Chinese
OpenRouter 于 2026-09-05 收录该模型(收录时间不等于模型正式发布日期);上下文:1,050,000 tokens;输入价:$10.0000/百万 tokens;输出价:$50.0000/百万 tokens;定位(来源描述):GPT-6 Astra is OpenAI's flagship model for demanding end-to-end work.
Full text
OpenAI: GPT-6 Astra
openai/gpt-6-astra
Compare
GPT-6 Astra is OpenAI's flagship model for demanding end-to-end work. It is suited for advanced analysis, software engineering, deep research, scientific work, and document creation, with particular strengths in long-horizon agentic tasks that involve computer and browser use.
Modalities
In / out price
$10 / $50per 1M
Context
1.1M
Released
Sep 4, 2026
Compare
Providers
Different companies host the same model. OpenRouter routes your request to one of them based on the routing mode you pick: Balanced (price + speed), Nitro (fastest), or Floor (cheapest).
Latency / throughput
$10.00$50.00$1.005.89s47 tok/s
96.66%
$10.00$50.00$1.005.67s32 tok/s
99.75%
$10.00$50.00$1.004.45s33 tok/s
99.96%
Not used in Standard routing:Why these endpoints are not used
Flex
$5.00$25.00$0.503.27s49 tok/s
100.00%
$11.00$55.00$1.101.48s30 tok/s
100.00%
$11.00$55.00$1.103.52s42 tok/s
100.00%
Fast
$20.00$100.00$2.001.65s71 tok/s
100.00%
Ultrafast
$60.00$300.00$6.00----
100.00%
ProviderInput / Output /M
Input $10.00 / Output $50.00
47 tok/s96.66% uptime
Input $10.00 / Output $50.00
32 tok/s99.75% uptime
Input $10.00 / Output $50.00
33 tok/s99.96% uptime
Not used in Standard routing:Why these endpoints are not used
Input $5.00 / Output $25.00
Flex
49 tok/s100.00% uptime
Input $11.00 / Output $55.00
30 tok/s100.00% uptime
Input $11.00 / Output $55.00
42 tok/s100.00% uptime
Input $20.00 / Output $100.00
Fast
71 tok/s100.00% uptime
Input $60.00 / Output $300.00
Ultrafast
-- tok/s100.00% uptime
Pricing
The average price customers actually pay for this model, next to the prices providers post. Caching and discounts mean the price actually paid is often well below the listed one.
Weighted average
Input
$1.882/M tokens
Output
$50.44/M tokens
GPT-6 Astra — Price history
Price history
No effective price data for this selection
$1.76$50.5893.9%22.4%
$1.90$50.8194.7%70.9%
$2.52$50.1390.3%2.8%
Not used in Standard routing:Why these endpoints are not used
Flex
$1.07$25.0089.8%2.3%
$2.21$58.8796.1%0.8%
Fast
$3.50$100.0093.5%0.9%
$13.09$55.000.0%0.0%
Ultrafast
————
ProviderInput / Output /M
Token share
Effective in $1.76 / Effective out $50.58
Token share
22.4%
Effective in $1.90 / Effective out $50.81
Token share
70.9%
Effective in $2.52 / Effective out $50.13
Token share
2.8%
Not used in Standard routing:Why these endpoints are not used
Effective in $1.07 / Effective out $25.00
FlexToken share
2.3%
Effective in $2.21 / Effective out $58.87
Token share
0.8%
Effective in $3.50 / Effective out $100.00
FastToken share
0.9%
Effective in $13.09 / Effective out $55.00
Token share
0.0%
Effective in — / Effective out —
UltrafastToken share —
Performance
Throughput is how fast the model writes (tokens per second — higher is better). Latency is total round-trip time (lower is better). TTFT is time-to-first-token — how long before you see anything appear (lower is better).
Throughput
71tok/s
P50, best across providers
Latency
1.48s
P50, best provider
Reasoning effortLatency / throughput
AutoExacto benchmarks
GPQA DiamondTAU-BenchAmazon Bedrock94.8%68.0%
Uptime
Uptime is the percentage of the past 3 days that at least one provider was responding to requests. Availability is the percentage of time that inference was successfully served. OpenRouter continuously monitors and uses the next-best provider when one returns an error.
Uptime (3d)The model was reachable. Request routed to a provider.
99.99%
Availability (3d)The model returned inference from any provider. Errors and empty responses count against it.
98.00%
Availability over the last 3 days
Last 72 hours
Availability 98.00%
3 Days Ago2 Days AgoYesterdayNow
Availability over the last 24 hours
OpenRouter Availability
95.87%
Without Routing
86.33%
When an error occurs in an upstream provider, we can recover by routing to another healthy provider, if your request filters allow it. You can access per-provider uptime data programmatically through the Endpoints API. Learn more about our load balancing and provider routing options.
Benchmarks
Scores on standardized evaluations and this model’s own outputs on shared benchmark prompts. Higher percentages are better — and each rank shows where this model lands among the models evaluated on that source.
Benchmark score summary for OpenAI: GPT-6 Astra (Artificial Analysis and Vals AI)
SourceBenchmarkScore
Artificial AnalysisGPT-6 Astra (High) Intelligence Index54.8
Artificial AnalysisGPT-6 Astra (High) Coding Index77.1
Artificial AnalysisGPT-6 Astra (High) Agentic Index48.2
Artificial AnalysisGPT-6 Astra (High) GPQA Diamond94.9%
Artificial AnalysisGPT-6 Astra (High) HLE53.1%
Artificial AnalysisGPT-6 Astra (High) AA-LCR80.0%
Artificial AnalysisGPT-6 Astra (High) τ-Bench Banking40.0%
Artificial AnalysisGPT-6 Astra (High) GDPval-AA51.9%
Artificial AnalysisGPT-6 Astra (High) CritPt28.9%
Artificial AnalysisGPT-6 Astra (High) SciCode55.4%
Artificial AnalysisGPT-6 Astra (High) Terminal-Bench 4.054.0%
Artificial AnalysisGPT-6 Astra (High) Terminal-Bench 2.189.9%
Artificial AnalysisGPT-6 Astra (High) AA-Omniscience Accuracy61.1%
Artificial AnalysisGPT-6 Astra (High) AA-Omniscience Non-Hallucination Rate57.5%
Artificial AnalysisGPT-6 Astra (Medium) Intelligence Index51.6
Artificial AnalysisGPT-6 Astra (Medium) Coding Index76.7
Artificial AnalysisGPT-6 Astra (Medium) Agentic Index46.0
Artificial AnalysisGPT-6 Astra (Medium) GPQA Diamond93.9%
Artificial AnalysisGPT-6 Astra (Medium) HLE52.7%
Artificial AnalysisGPT-6 Astra (Medium) AA-LCR79.7%
Artificial AnalysisGPT-6 Astra (Medium) τ-Bench Banking35.5%
Artificial AnalysisGPT-6 Astra (Medium) GDPval-AA50.1%
Artificial AnalysisGPT-6 Astra (Medium) CritPt29.1%
Artificial AnalysisGPT-6 Astra (Medium) SciCode54.2%
Artificial AnalysisGPT-6 Astra (Medium) Terminal-Bench 4.049.5%
Artificial AnalysisGPT-6 Astra (Medium) Terminal-Bench 2.189.5%
Artificial AnalysisGPT-6 Astra (Medium) AA-Omniscience Accuracy59.6%
Artificial AnalysisGPT-6 Astra (Medium) AA-Omniscience Non-Hallucination Rate54.8%
Artificial AnalysisGPT-6 Astra (Xhigh) Intelligence Index55.3
Artificial AnalysisGPT-6 Astra (Xhigh) Coding Index75.9
Artificial AnalysisGPT-6 Astra (Xhigh) Agentic Index50.2
Artificial AnalysisGPT-6 Astra (Xhigh) GPQA Diamond96.3%
Artificial AnalysisGPT-6 Astra (Xhigh) HLE54.6%
Artificial AnalysisGPT-6 Astra (Xhigh) AA-LCR80.0%
Artificial AnalysisGPT-6 Astra (Xhigh) τ-Bench Banking43.1%
Artificial AnalysisGPT-6 Astra (Xhigh) GDPval-AA53.6%
Artificial AnalysisGPT-6 Astra (Xhigh) CritPt31.4%
Artificial AnalysisGPT-6 Astra (Xhigh) SciCode55.7%
Artificial AnalysisGPT-6 Astra (Xhigh) Terminal-Bench 4.059.6%
Artificial AnalysisGPT-6 Astra (Xhigh) Terminal-Bench 2.189.1%
Artificial AnalysisGPT-6 Astra (Xhigh) AA-Omniscience Accuracy61.0%
Artificial AnalysisGPT-6 Astra (Xhigh) AA-Omniscience Non-Hallucination Rate55.4%
Artificial AnalysisGPT-6 Astra (Low) Intelligence Index43.3
Artificial AnalysisGPT-6 Astra (Low) Coding Index75.7
Artificial AnalysisGPT-6 Astra (Low) Agentic Index38.8
Artificial AnalysisGPT-6 Astra (Low) GPQA Diamond93.1%
Artificial AnalysisGPT-6 Astra (Low) HLE49.2%
Artificial AnalysisGPT-6 Astra (Low) AA-LCR80.0%
Artificial AnalysisGPT-6 Astra (Low) τ-Bench Banking32.0%
Artificial AnalysisGPT-6 Astra (Low) GDPval-AA45.2%
Artificial AnalysisGPT-6 Astra (Low) CritPt26.3%
Artificial AnalysisGPT-6 Astra (Low) SciCode54.1%
Artificial AnalysisGPT-6 Astra (Low) Terminal-Bench 4.041.9%
Artificial AnalysisGPT-6 Astra (Low) Terminal-Bench 2.188.0%
Artificial AnalysisGPT-6 Astra (Low) AA-Omniscience Accuracy58.6%
Artificial AnalysisGPT-6 Astra (Low) AA-Omniscience Non-Hallucination Rate55.8%
Artificial AnalysisGPT-6 Astra (Max) Intelligence Index56.9
Artificial AnalysisGPT-6 Astra (Max) Coding Index76.9
Artificial AnalysisGPT-6 Astra (Max) Agentic Index51.0
Artificial AnalysisGPT-6 Astra (Max) GPQA Diamond96.1%
Artificial AnalysisGPT-6 Astra (Max) HLE54.7%
Artificial AnalysisGPT-6 Astra (Max) AA-LCR80.7%
Artificial AnalysisGPT-6 Astra (Max) τ-Bench Banking41.4%
Artificial AnalysisGPT-6 Astra (Max) GDPval-AA54.3%
Artificial AnalysisGPT-6 Astra (Max) CritPt31.7%
Artificial AnalysisGPT-6 Astra (Max) SciCode56.5%
Artificial AnalysisGPT-6 Astra (Max) Terminal-Bench 4.059.1%
Artificial AnalysisGPT-6 Astra (Max) Terminal-Bench 2.188.4%
Artificial AnalysisGPT-6 Astra (Max) AA-Omniscience Accuracy61.6%
Artificial AnalysisGPT-6 Astra (Max) AA-Omniscience Non-Hallucination Rate50.2%
Vals AIGPT-6 Astra BioMysteryBench79.3%
Vals AIGPT-6 Astra Code Migration v1.180.6%
Vals AIGPT-6 Astra CyberBench v1.141.1%
Vals AIGPT-6 Astra EMB v1.172.7%
Vals AIGPT-6 Astra Finance Agent v2.160.4%
Vals AIGPT-6 Astra Vals Finance Index55.6%
Vals AIGPT-6 Astra Financial Coaching Bench67.5%
Vals AIGPT-6 Astra HLAB-Vals-1.17.1%
Vals AIGPT-6 Astra Immigration67.7%
Vals AIGPT-6 Astra IOI100.0%
Vals AIGPT-6 Astra Vals Legal Index68.4%
Vals AIGPT-6 Astra Legal Research Bench39.4%
Vals AIGPT-6 Astra MedCode48.5%
Vals AIGPT-6 Astra MedScribe87.9%
Vals AIGPT-6 Astra MysteryMechanism53.2%
Vals AIGPT-6 Astra ProgramBench5.5%
Vals AIGPT-6 Astra ProofBench v1.199.0%
Vals AIGPT-6 Astra Robo-bench23.3%
Vals AIGPT-6 Astra Vals RSI Index27.1%
Vals AIGPT-6 Astra SAGE v146.4%
Vals AIGPT-6 Astra Vals Science Index65.1%
Vals AIGPT-6 Astra SRE Bench56.9%
Vals AIGPT-6 Astra SWE-bench Multimodal43.3%
Vals AIGPT-6 Astra Tax Agent Bench20.7%
Vals AIGPT-6 Astra Terminal-Bench 4.059.6%
Vals AIGPT-6 Astra Terminal-Bench-Science62.9%
Vals AIGPT-6 Astra Vals Index65.0%
Vals AIGPT-6 Astra Vibe Code Bench 1-10027.6%
Vals AIGPT-6 Astra Vibe Code Bench v1.189.6%
Apps
Public apps that send the most traffic to this model. Good signal for what real production workloads look like — and a hint at which use cases this model is best suited for.
1.
Codex
A coding agent that helps you build and ship with AI
478Btokens
2.
Hermes Agent
Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents.
50.2Btokens
3.
Framer
Create a professional website with Framer’s no-code AI website builder. Design freely, manage CMS content, optimize SEO, collaborate, and publish fast.
49.7Btokens
4.
Claude Code
Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans and executes changes across files, runs tests, and iterates on failures, all from natural language prompts.
30Btokens
5.
pi
There are many coding agents, but this one is yours.
27.6Btokens
Rest of OpenRouter
7.01B
Top 5 Apps
1.91B
Activity
Token volume and request traffic to this model over time.
API
Drop-in code to call this model. OpenRouter's API is OpenAI-compatible — most SDKs work by just swapping the base URL. The only thing that changes between models is the model slug below.
OpenAI: GPT Astra versions
Not currently available through EU in-region routing. In-region routing docs
Documentation
Guides, cookbooks, blog posts, and tools for building with OpenAI: GPT-6 Astra on OpenRouter.
GPT-6 Migration GuideCookbook
Frequently asked questions
GPT-6 Astra is OpenAI's flagship model for demanding end-to-end work. It is suited for advanced analysis, software engineering, deep research, scientific work, and document creation, with particular strengths in long-horizon agentic tasks that involve computer and browser use.
The GPT-6 Astra documentation on OpenRouter includes the GPT-6 Migration Guide cookbook. The quickstart covers API keys and your first request, and the API reference lists every parameter.
GPT-6 Astra costs $10.00/M input tokens and $50.00/M output tokens, with separate rates for Cache Read at $1.00/M tokens and Cache Write at $12.50/M tokens.
GPT-6 Astra has a 1,050,000 token context window. It supports up to 128,000 completion tokens.
Yes. GPT-6 Astra accepts tools and tool_choice for function calling. It also supports structured outputs via a JSON schema in response_format.
Yes. OpenAI caches prompts of 1024 tokens or more automatically, and GPT-6 Astra bills a cache write on each cached prompt whether or not the cached prefix is read again. To turn caching off for a request, set prompt_cache_options.mode to "explicit" at the request root and do not mark any content block with prompt_cache_breakpoint. Explicit mode disables OpenAI-managed breakpoints, so a request with none of your own has no cacheable prefix and is billed at the plain input rate. See the prompt caching guide.
GPT-6 Astra accepts files such as PDFs, images, and text as input and returns text.
GPT-6 Astra is served by 3 providers on OpenRouter: OpenAI, Amazon Bedrock, and Azure. Requests are routed to the best available provider, with automatic failover to the others, and you can pin or exclude providers with provider routing.
GPT-6 Astra was released on September 4, 2026.