OpenAIproprietary
Released: 2026-09-03GPT-6 Astra
OpenAI sixth-generation Astra flagship combining deep test-time compute with continuous test verification.
Intelligence Index
94.2/ 100
Calibrated multi-domain compositeThroughput Speed
62tok/sec
Output streaming throughputTime To First Token
420ms
Initial chunk server latencyLiveBench Contam-Free
84.2%
September 2026 suite scoreLiveBench Multi-Domain Evaluation Breakdown
Monthly refreshed contamination-free test setsLogical Reasoning91.7%
Coding & Repo Repair85.9%
Mathematics (AIME)90.2%
Data Analysis & Tables81.4%
Language & Comprehension77.8%
Instruction Following78.0%
Sourced Benchmark Evaluations (61)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| ARC-AGI-3 | Active | 59.6% | 93 | 2026-09-01 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 71.5% | 75 | 2026-08-25 | independent | Source ↗ |
| FrontierMath | Active | 26.5% | 98 | 2026-08-22 | independent | Source ↗ |
| PutnamBench | Active | 49.1% | 90 | 2026-08-21 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 61.7% | 20 | 2026-08-19 | independent | Source ↗ |
| Omni-MATH | Deprecated | 69.9% | 10 | 2026-08-19 | independent | Source ↗ |
| Soohak | Active | 71.8% | 98 | 2026-08-17 | independent | Source ↗ |
| FrontierCode | Active | 52.7% | 98 | 2026-08-17 | independent | Source ↗ |
| Humanity's Last Exam | Active | 40.3% | 93 | 2026-08-15 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 66.3% | 75 | 2026-08-15 | independent | Source ↗ |
| AIME | Nearing Saturation | 89.8% | 75 | 2026-08-15 | independent | Source ↗ |
| OlympiadBench | Active | 77.5% | 80 | 2026-08-14 | independent | Source ↗ |
| MathArena | Active | 75.6% | 99 | 2026-08-13 | independent | Source ↗ |
| FireBench | Active | 65.5% | 95 | 2026-08-13 | independent | Source ↗ |
| Terminal-Bench | Active | 63.6% | 80 | 2026-08-13 | independent | Source ↗ |
| OpenAI-MRCR | Active | 79.1% | 80 | 2026-08-12 | independent | Source ↗ |
| SimpleQA | Active | 49% | 90 | 2026-08-12 | independent | Source ↗ |
| CodeContests | Deprecated | 59% | 10 | 2026-08-12 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 64% | 80 | 2026-08-11 | independent | Source ↗ |
| SWE-bench Pro | Active | 56.3% | 90 | 2026-08-11 | independent | Source ↗ |
| LongBench-Pro | Active | 77.2% | 80 | 2026-08-11 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 75.4% | 78 | 2026-08-10 | independent | Source ↗ |
| MultiPL-E | Deprecated | 83.5% | 10 | 2026-08-10 | independent | Source ↗ |
| LiveCodeBench | Saturated | 74.5% | 39 | 2026-08-09 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 53% | 80 | 2026-08-09 | independent | Source ↗ |
| MATH | Saturated | 92.6% | 25 | 2026-08-09 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 80.1% | 65 | 2026-08-09 | independent | Source ↗ |
| Mercury | Deprecated | 79.9% | 10 | 2026-08-08 | independent | Source ↗ |
| SCROLLS | Nearing Saturation | 75.4% | 65 | 2026-08-08 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 86.7% | 75 | 2026-08-08 | independent | Source ↗ |
| RULER | Nearing Saturation | 86.7% | 84 | 2026-08-07 | independent | Source ↗ |
| TabMWP | Saturated | 86.9% | 25 | 2026-08-07 | independent | Source ↗ |
| CyberSecEval | Deprecated | 81.7% | 10 | 2026-08-07 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 70.8% | 25 | 2026-08-06 | independent | Source ↗ |
| MMLU-Redux | Saturated | 92.3% | 25 | 2026-08-06 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 82.9% | 78 | 2026-08-06 | independent | Source ↗ |
| LongBench | Nearing Saturation | 81% | 65 | 2026-08-05 | independent | Source ↗ |
| SWE-bench | Deprecated | 61.7% | 10 | 2026-08-05 | independent | Source ↗ |
| MGSM | Nearing Saturation | 89.8% | 65 | 2026-08-05 | independent | Source ↗ |
| AGIEval | Saturated | 78.9% | 25 | 2026-08-04 | independent | Source ↗ |
| QASPER | Nearing Saturation | 78.2% | 65 | 2026-08-04 | independent | Source ↗ |
| TruthfulQA | Saturated | 82.9% | 25 | 2026-08-04 | independent | Source ↗ |
| GSM-Hard | Deprecated | 90.7% | 10 | 2026-08-03 | independent | Source ↗ |
| SuperGLUE | Saturated | 93.5% | 25 | 2026-08-03 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 85.4% | 23 | 2026-08-02 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 8% | 75 | 2026-08-02 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 89.9% | 25 | 2026-08-02 | independent | Source ↗ |
| WinoGrande | Saturated | 95.4% | 28 | 2026-08-01 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 92.6% | 65 | 2026-08-01 | independent | Source ↗ |
| IFEval | Deprecated | 76.4% | 25 | 2026-08-01 | independent | Source ↗ |
| GPQA | Nearing Saturation | 82.9% | 78 | 2026-08-01 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 98.8% | 44 | 2026-07-31 | independent | Source ↗ |
| HumanEval+ | Saturated | 98.1% | 25 | 2026-07-31 | independent | Source ↗ |
| CommonsenseQA | Saturated | 96.3% | 28 | 2026-07-31 | independent | Source ↗ |
| MBPP | Saturated | 96.2% | 25 | 2026-07-30 | independent | Source ↗ |
| GSM8K | Saturated | 100% | 25 | 2026-07-30 | independent | Source ↗ |
| GLUE | Saturated | 99% | 28 | 2026-07-30 | independent | Source ↗ |
| MMLU | Saturated | 96.1% | 25 | 2026-07-29 | independent | Source ↗ |
| HumanEval | Saturated | 100% | 25 | 2026-07-29 | independent | Source ↗ |
| ARC-AGI | Saturated | 86.2% | 25 | 2026-07-29 | independent | Source ↗ |
| LiveBench (Continuous) | Active | 84.2% | 94 | 2025-02-01 | independent | Source ↗ |
OpenAI Compatible API ExecutionAPI Ready
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="gpt-6-astra",
messages=[{"role": "user", "content": "Evaluate multi-step logic problem"}],
temperature=0.2
)
print(response.choices[0].message.content)Evaluation & Sourcing Notes
Scores listed for GPT-6 Astra represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.