GPT-6 Astra
OpenAI's GPT-6 flagship at $10/$50 per 1M tokens, alongside the GPT-6.1 Sol and GPT-6 Luna tiers, whose prices OpenAI cut by 50% or more.
Model Story
GPT-6 Astra was released by OpenAI on September 3, 2026, as the flagship of the GPT-6 family. Artificial Analysis lists it at 53 on its Intelligence Index v4.3.x, level with Claude Fable 5.1 and Gemini 4 Argon and behind Claude Opus 5.5 (58 at max effort). It completes an Intelligence Index task for $3.26 on average, less than Fable 5.1's $7.63 but more than Gemini 4 Argon's $1.99 at promotional prices.
Astra continues a clear trend at the top of OpenAI's lineup. GPT-5.4 listed at $2.50 input / $15 output per 1M tokens and GPT-5.5 at $5 / $30; GPT-6 Astra lists at $10 input / $50 output. From GPT-5.4 to GPT-5.5 to GPT-6 Astra, the flagship price has risen at every step.
Below the flagship, prices went the other way. GPT-6.1 Sol costs $2 input / $10 output per 1M tokens and scores 52 on the Artificial Analysis index, just one point behind Astra. GPT-6 Luna, the budget tier, costs $0.10 input / $0.50 output. OpenAI cut its mid-tier and budget prices by 50% or more and says the Sol and Luna prices are permanent, not promotional.
For Faraday Machines customers, GPT-6 models are typically a hybrid component: local open-weight models handle sensitive internal data, while specific tasks are routed to OpenAI via API.
Key Specifications
| Developer | OpenAI |
|---|---|
| Release Date | September 3, 2026 |
| Architecture | Undisclosed (proprietary) |
| License | Proprietary (API / enterprise license) |
| Family Tiers | GPT-6 Astra (flagship), GPT-6.1 Sol (mid-tier), GPT-6 Luna (budget) |
| Cached Input | $1.00 per 1M tokens |
| Previous Flagship | GPT-5.5 ($5 / $30 per 1M tokens) |
| Also Available | GPT-5.6 Sol and GPT-5.6 Terra |
API Pricing
GPT-6 Family Pricing
| GPT-6 Astra | $10 input / $50 output per 1M tokens; AA Intelligence Index v4.3.x: 53 |
|---|---|
| GPT-6.1 Sol | $2 input / $10 output per 1M tokens; AA Intelligence Index v4.3.x: 52 |
| GPT-6 Luna | $0.10 input / $0.50 output per 1M tokens |
OpenAI's mid-tier and budget cuts are real, and OpenAI says they are permanent. But the flagship price rose again, and cloud prices elsewhere move in both directions without warning: Gemini 4 Argon launched at a promotional 50%-off rate that will expire. Whatever the rate card says, per-token bills scale with every query your team runs.
Benchmarks
All scores above are independent Artificial Analysis results on the v4.3 index, introduced in September 2026. Every model scored lower on v4.3 than on the earlier v4.1 index, so these figures are not comparable with v4.1 scores on older profiles such as GPT-5.4.
On-Premises Considerations
GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are fully proprietary models with no open-weight release, so none of them can be self-hosted on Faraday Machines hardware. Faraday supports hybrid deployment patterns where sensitive data is processed locally by open-weight models, with sanitized or abstracted queries sent to OpenAI only where its capabilities are needed.
For work that has to stay on-premises, the open-weight models Faraday ships cover each tier: Qwen3.8-27B on Starter; GLM-5.3-Flash and MiMo-V2.6-Flash on Growth; and MiniMax M3 and DeepSeek V4.1 Flash on Scale. On larger clusters, MiMo-V2.6-Pro scores 46 on the Artificial Analysis Intelligence Index v4.3.2, about 7 points behind GPT-6 Astra, at roughly $0.13 per task against Astra's $3.26. GLM-5.3 and Kimi K3 are also available on larger clusters.
For organizations evaluating total cost of ownership, Faraday's AI cost calculator compares API costs with on-premises deployment, making it easy to weigh hybrid architectures against pure cloud or pure on-premises approaches.
Frequently Asked Questions
Can I run GPT-6 Astra on-premises?
No. GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are proprietary OpenAI models with no open-weight release, so none of them can be self-hosted on Faraday Machines hardware. Faraday Machines supports hybrid setups where sensitive data is processed locally by open-weight models such as DeepSeek V4.1 Flash or GLM-5.3-Flash, with sanitized queries sent to OpenAI only where its capabilities are needed.
How much does GPT-6 Astra cost?
GPT-6 Astra lists at $10 per 1M input tokens and $50 per 1M output tokens, with cached input at $1 per 1M. That is up from GPT-5.5's $5 / $30 and GPT-5.4's $2.50 / $15. Artificial Analysis measures an average of $3.26 per Intelligence Index task. OpenAI's lower tiers got cheaper: GPT-6.1 Sol costs $2 / $10 and GPT-6 Luna $0.10 / $0.50.
Is GPT-6.1 Sol almost as good as GPT-6 Astra?
Nearly, on the Artificial Analysis Intelligence Index v4.3.x: GPT-6.1 Sol scores 52, one point behind GPT-6 Astra's 53, at a fifth of the per-token price ($2 / $10 versus $10 / $50 per 1M tokens). OpenAI cut its mid-tier and budget prices by 50% or more and says the Sol and Luna prices are permanent, not promotional.
How does GPT-6 Astra compare to Claude Opus 5.5 and open-weight models?
GPT-6 Astra scores 53 on the Artificial Analysis Intelligence Index v4.3.x, level with Claude Fable 5.1 and Gemini 4 Argon and behind Claude Opus 5.5 at 58 (max effort). The top open-weight model, MiMo-V2.6-Pro, scores 46 on v4.3.2, about 7 points behind Astra, at roughly $0.13 per task against Astra's $3.26.
Explore Other Models
See how every model compares on price, benchmarks and hardware fit in the AI Model Comparison.