Reasoning
198-question Diamond subset; independent AA evaluation
- 01
- 02
- 03
- 03
- 03
GPT-6 Astra
Across the task benchmarks we track, GPT-6 Astra’s strongest relative results are in reasoning, coding, and web research. On GPQA Diamond, it scores 96.3%, ranking #1 among 30 models with reported results in our dataset.
Updated: Data as of:
🥇
Strongest results
| Benchmark | Measures | Score |
|---|---|---|
| Agent Index ↗v1.2 Preview | Overall agent capability | 87.6 / 100#2 of 26 |
| GDP.pdf ↗100 tasks / All-pass / five attempts | Professional document reasoning | 32.2%#1 of 55 |
| GPQA ↗Diamond / Artificial Analysis | Reasoning | 96.26%#1 of 30 |
| DeepSWE ↗v1.1 | Coding agents | 74%Joint #1 of 11 |
Model details
| Model | GPT-6 Astra |
|---|---|
| Agent Index v1.2 | 87.6 / 100 · Rank 2 · Preview |
| Creator | OpenAI (opens in a new tab) |
| Availability | Proprietary |
| Context size | 1,000,000 tokens (opens in a new tab) |
| Max output | 128,000 tokens (opens in a new tab) |
| Knowledge cutoff | 2026-04-30 |
| License | Not reported |
Pricing and performance
Per provider and reported configuration. Prices are USD per 1M tokens.
| Provider / configuration | Contexttokens | InputUSD / 1M tokens | OutputUSD / 1M tokens | LatencyTTFT · seconds | Speedtokens / sec |
|---|---|---|---|---|---|
| GPT-6 Astra (max) | 524,288 | $10.00 (opens in a new tab) | $50.00 (opens in a new tab) | 302.232 (opens in a new tab) | 89.577 (opens in a new tab) |
| GPT-6 Astra (max) | 524,288 | $10.00 (opens in a new tab) | $50.00 (opens in a new tab) | 349.1 (opens in a new tab) | 69.338 (opens in a new tab) |
| GPT-6 Astra (max) | 1,000,000 | $10.00 (opens in a new tab) | $50.00 (opens in a new tab) | 365.127 (opens in a new tab) | 70.564 (opens in a new tab) |
| GPT-6 Astra (max) (Fast) | 1,000,000 | $20.00 (opens in a new tab) | $100.00 (opens in a new tab) | Not reported | Not reported |
| GPT-6 Astra (max) (Flex) | 1,000,000 | $5.00 (opens in a new tab) | $25.00 (opens in a new tab) | 322.971 (opens in a new tab) | 53.896 (opens in a new tab) |
| GPT-6 Astra (max) (Ultrafast) | 1,000,000 | $60.00 (opens in a new tab) | $300.00 (opens in a new tab) | Not reported | Not reported |
Benchmark performance
GPT-6 Astra in color, alongside leading models. Latest stored reports, best effort per report date.
GPQA Diamond ↗
Coding
113 tasks; mini-swe-agent
- 01
- 01
- 01
- 04
- 05
Terminal Use
Official 4.0 board; native agent harnesses
- 01
- 02
- 03
- 03
- 05
Web Research
Professional benchmark table; OpenAI Astra release comparison
- 01
- 02
- 03
- 04
Work Automations
Private held-out set; API mode; 50 steps; strict pass/fail
- 01
- 02
- 03
- 04
- 05
Knowledge Work
220 tasks; Stirrup agent; pairwise Elo
- 01
- 02
- 03
- 04
- 05
- 17
Reported benchmarks
All published results in our dataset, with versions, evaluation settings, and sources.
| Benchmark | Score | Evaluation setup | Source |
|---|---|---|---|
| AA-Briefcase1.1 / combined EloCombined Elo · Higher is better | 1,569 Elo | selection: publisher default displayed configuration · GPT-6 Astra (max) | Artificial Analysis (opens in a new tab)independent |
| AutomationBench1.0.6 / Zapier held-out private setStrict task completion · Higher is better | 25.27% | GPT-6 Astra (None); private held-out 657 tasks; API mode; 50-step cap; strict completion | Zapier ↗ (opens in a new tab)benchmark publisher |
| AutomationBench1.0.6 / Zapier held-out private setStrict task completion · Higher is better | 30.29% | GPT-6 Astra (Low); private held-out 657 tasks; API mode; 50-step cap; strict completion | Zapier ↗ (opens in a new tab)benchmark publisher |
| AutomationBench1.0.6 / Zapier held-out private setStrict task completion · Higher is better | 34.09% | GPT-6 Astra (Medium); private held-out 657 tasks; API mode; 50-step cap; strict completion | Zapier ↗ (opens in a new tab)benchmark publisher |
| AutomationBench1.0.6 / Zapier held-out private setStrict task completion · Higher is better | 37.14% | GPT-6 Astra (High); private held-out 657 tasks; API mode; 50-step cap; strict completion | Zapier ↗ (opens in a new tab)benchmark publisher |
| AutomationBench1.0.6 / Zapier held-out private setStrict task completion · Higher is better | 38.96% | GPT-6 Astra (XHigh); private held-out 657 tasks; API mode; 50-step cap; strict completion | Zapier ↗ (opens in a new tab)benchmark publisher |
| AutomationBench1.0.6 / Zapier held-out private setStrict task completion · Higher is better | 41.4% | GPT-6 Astra (Max); private held-out 657 tasks; API mode; 50-step cap; strict completion | Zapier ↗ (opens in a new tab)benchmark publisher |
| AutomationBench-AA1.0.6 / AA private 657Objectives completed without guardrail violations · Higher is better | 68.5% | reported precision decimals: 1 · result quality: reported · GPT-6 Astra (max); private held-out 657 tasks; 50 turns; objective completion with zero credit for any guardrail violation. · source model href: /models/gpt-6-astra · task count: 657 · Tools enabled · turn cap: 50 | Artificial Analysis (opens in a new tab)independent |
| BrowseCompKimi K3.1 comparison / 200K compactionAccuracy · Higher is better | 91.5% | 200K context compaction; max 500 steps; temperature 1; top-p 1 | Moonshot AI ↗ (opens in a new tab)provider reported |
| BrowseCompOpenAI Astra launch comparison / September 2026Accuracy · Higher is better | 91.5% | OpenAI GPT-6 Astra launch comparison; best reported reasoning setting; see source evaluation footnotes | OpenAI ↗ (opens in a new tab)provider reported |
| DeepSWEv1.1Pass@1 · Higher is better | 74% | mini-swe-agent; effort=xhigh; ±3 percentage points | Datacurve ↗ (opens in a new tab)benchmark publisher |
| DeepSWEv1.1Pass@1 · Higher is better | 74% | xhigh; ±3 percentage points | Datacurve ↗ (opens in a new tab)benchmark publisher |
| DeepSWEv1.1 / OpenAI Astra launch comparisonPass@1 · Higher is better | 74.1% | OpenAI GPT-6 Astra launch comparison; best reported reasoning setting; see source evaluation footnotes | OpenAI ↗ (opens in a new tab)provider reported |
| GDP.pdf100 tasks / All-pass / five attemptsAccuracy · Higher is better | 32.2% | selection: best reported configuration per release · GPT-6 Astra (xhigh) · tools: benchmark-defined | Artificial Analysis (opens in a new tab)independent |
| GDP.pdf100 tasks / All-pass / five attemptsAccuracy · Higher is better | 31% | selection: publisher default displayed configuration · GPT-6 Astra (max) | Artificial Analysis (opens in a new tab)independent |
| GDPval-AA2.1Elo · Higher is better | 1,498.47 Elo | GPT-6 Astra (Medium); 220 tasks; Stirrup agent; pairwise Elo | Artificial Analysis (opens in a new tab)independent |
| GDPval-AA2.1Elo · Higher is better | 1,536.04 Elo | GPT-6 Astra (High); 220 tasks; Stirrup agent; pairwise Elo | Artificial Analysis (opens in a new tab)independent |
| GDPval-AA2.1Elo · Higher is better | 1,403.26 Elo | GPT-6 Astra (Low); 220 tasks; Stirrup agent; pairwise Elo | Artificial Analysis (opens in a new tab)independent |
| GDPval-AA2.1Elo · Higher is better | 1,583.69 Elo | GPT-6 Astra (Max); 220 tasks; Stirrup agent; pairwise Elo | Artificial Analysis (opens in a new tab)independent |
| GDPval-AA2.1Elo · Higher is better | 1,569.36 Elo | GPT-6 Astra (Xhigh); 220 tasks; Stirrup agent; pairwise Elo | Artificial Analysis (opens in a new tab)independent |
| GPQADiamond / Artificial AnalysisAccuracy · Higher is better | 93.939% | GPT-6 Astra (Medium); 198-question Diamond subset; independent AA evaluation | Artificial Analysis (opens in a new tab)independent |
| GPQADiamond / Artificial AnalysisAccuracy · Higher is better | 94.949% | GPT-6 Astra (High); 198-question Diamond subset; independent AA evaluation | Artificial Analysis (opens in a new tab)independent |
| GPQADiamond / Artificial AnalysisAccuracy · Higher is better | 93.131% | GPT-6 Astra (Low); 198-question Diamond subset; independent AA evaluation | Artificial Analysis (opens in a new tab)independent |
| GPQADiamond / Artificial AnalysisAccuracy · Higher is better | 96.061% | GPT-6 Astra (Max); 198-question Diamond subset; independent AA evaluation | Artificial Analysis (opens in a new tab)independent |
| GPQADiamond / Artificial AnalysisAccuracy · Higher is better | 96.263% | GPT-6 Astra (Xhigh); 198-question Diamond subset; independent AA evaluation | Artificial Analysis (opens in a new tab)independent |
| HLE-Diamond1,000 questions / no toolsAccuracy · Higher is better | 63.8% | selection: best reported configuration per release · GPT-6 Astra (Max) · No tools | Artificial Analysis (opens in a new tab)independent |
| Humanity's Last ExamStable HLE — full setAccuracy · Higher is better | 57.2% | OpenAI Astra launch comparison; reported with tools. Full-set interpretation corroborated by Moonshot’s full-set comparison. · subset: full · Tools enabled | OpenAI ↗ (opens in a new tab)provider reported |
| Humanity's Last ExamStable HLE — full setAccuracy · Higher is better | 57.2% | Max effort as labeled in Moonshot comparison; GPT models may include cyberguards. Confirms OpenAI launch table HLE with tools 57.2. · subset: full · Tools enabled | Moonshot AI ↗ (opens in a new tab)provider reported |
| Humanity's Last ExamStable HLE — text-only subsetAccuracy · Higher is better | 52.734% | GPT-6 Astra (Medium); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools | Artificial Analysis (opens in a new tab)independent |
| Humanity's Last ExamStable HLE — text-only subsetAccuracy · Higher is better | 53.058% | GPT-6 Astra (High); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools | Artificial Analysis (opens in a new tab)independent |
| Humanity's Last ExamStable HLE — text-only subsetAccuracy · Higher is better | 49.212% | GPT-6 Astra (Low); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools | Artificial Analysis (opens in a new tab)independent |
| Humanity's Last ExamStable HLE — text-only subsetAccuracy · Higher is better | 54.68% | GPT-6 Astra (Max); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools | Artificial Analysis (opens in a new tab)independent |
| Humanity's Last ExamStable HLE — text-only subsetAccuracy · Higher is better | 54.588% | GPT-6 Astra (Xhigh); 2,158 text-only questions; no tools; original HLE equality checker; GPT-5.6 Luna medium judge · subset: text · No tools | Artificial Analysis (opens in a new tab)independent |
| LifeValNovember 2026Unweighted average reward · Higher is better | 80.6% | aggregation: Average attempts within each task, then average all 169 task means equally. No willingness-to-pay weighting. Reward multiplied by 100; published display values rounded to one decimal place. · Launch-blog xhigh setting · tasks: 169 | OpenAI ↗ (opens in a new tab)benchmark publisher · 2026-11-02 |
| LiveBench: Agentic2.0 / 2026-10-25Accuracy · Higher is better | 77.5% | release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: agentic | LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25 |
| LiveBench: Browser Use2.0 / 2026-10-25Accuracy · Higher is better | 91.3% | release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: browser-use | LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25 |
| LiveBench: Coding2.0 / 2026-10-25Accuracy · Higher is better | 93.7% | release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: coding | LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25 |
| LiveBench: Data Analysis2.0 / 2026-10-25Accuracy · Higher is better | 83% | release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: data-analysis | LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25 |
| LiveBench: Design2.0 / 2026-10-25Accuracy · Higher is better | 82% | release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: design | LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25 |
| LiveBench: Reasoning2.0 / 2026-10-25Accuracy · Higher is better | 92.7% | release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: reasoning | LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25 |
| LiveBench: Research2.0 / 2026-10-25Accuracy · Higher is better | 85.2% | release: 2026-10-25 · Highest supported effort unless publisher reports an exception; single run · suite: research | LiveBench ↗ (opens in a new tab)benchmark publisher · 2026-10-25 |
| Terminal-Bench4.0Task success · Higher is better | 58.2% | Codex; effort=max; 58.2% ± 2.8% | Terminal-Bench ↗ (opens in a new tab)benchmark publisher |
| Terminal-Bench4.0Task success · Higher is better | 58.2% | Codex; max; ±2.8 percentage points | Terminal-Bench ↗ (opens in a new tab)benchmark publisher |
| Terminal-Bench4.0 / mini-swe-agentTask success · Higher is better | 59.596% | harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-6 Astra (Xhigh) | Artificial Analysis (opens in a new tab)independent |
