Model comparison
GPT-6 Astra vs Claude Sonnet 5
GPT-6 Astra costs $10.00 / $50.00 and Claude Sonnet 5 $2.00 / $10.00 per 1M input / output tokens. Both accept about 1M tokens of context, text, images and files.
Which one to choose
Picked by a published rule, not by an editor. Scenarios without data are left out.
Everyday requests
Claude Sonnet 5
$0.035 vs $0.007 for 1,000 input and 500 output tokens.
Long prompts
Claude Sonnet 5
A 500K-token prompt with a 2K-token answer costs $10.15 on GPT-6 Astra and $1.02 on Claude Sonnet 5.
Context window
About the same
GPT-6 Astra: 1.05M tokens; Claude Sonnet 5: 1M. Differences under 10% count as a tie.
Mixed inputs
Claude Sonnet 5
GPT-6 Astra: Text, Images. Claude Sonnet 5: Text, Images, Files (PDF).
How the verdicts are calculated
- Everyday requests: lower cost for 1,000 input + 500 output tokens at standard rates.
- Long prompts: lower cost for a 500K-token prompt and 2K-token answer, including long-context surcharges.
- Context window: bigger window wins; differences under 10% count as a tie.
- Mixed inputs: more input types accepted; a tie is not shown until multimodal benchmark scores are collected.
Not shown yet
- Coding: needs SWE-bench Verified scores
- Speed: needs independent output-speed measurements
- Privacy: needs data-retention and training-use policies
Side by side
The better value in each row is marked. Observed from LiteLLM and OpenRouter, checked against vendor pricing pages.
| Spec | GPT-6 Astra | Claude Sonnet 5 |
|---|---|---|
| Input priceper 1M tokens | $10.00 | $2.00 (better) |
| Output priceper 1M tokens | $50.00 | $10.00 (better) |
| Cached inputper 1M tokens | $1.00 | $0.20 (better) |
| Typical request1,000 input + 500 output tokens | $0.035 | $0.007 (better) |
| Long-prompt pricingCompared on a 500K-token prompt | $20.00 / $75.00 above 272K tokens | No surcharge listed (better) |
| Batch priceinput / output per 1M tokens | $5.00 / $25.00 | $1.00 / $5.00 (better) |
| Context window | 1.05M tokens | 1M tokens |
| Max output | 128K tokens | 128K tokens |
| Knowledge cutoff | Apr 2026 | Not disclosed |
| Input types | Text, Images | Text, Images, Files (PDF) (better) |
| Reasoning | Adjustable effort | Adjustable effort |
| Tool use | Yes | Yes |
| Structured output (JSON) | Yes | Yes |
| BenchmarksGPQA Diamond, SWE-bench Verified, AIME 2025, LMArena | Not collected yet | Not collected yet |
| Output speedtokens per second | Not collected yet | Not collected yet |