| Claude Opus 5 | 33.12 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort high; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Claude Sonnet 5 | 8.76 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort high; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Claude Opus 5 | 17.9 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort low; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Claude Sonnet 5 | 4.03 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort low; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Claude Opus 5 | 33.54 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort max; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Claude Opus 5 | 24.13 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort medium; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Claude Sonnet 5 | 5.99 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort medium; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Claude Opus 5 | 35.91 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort xhigh; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Claude Sonnet 5 | 8.98 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort xhigh; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Qwen 3.8 Max | 47.69 | points | Native editable single-slide PPTX; 500 tasks; Claude Code; reasoning effort xhigh; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Luna | 19.47 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort high; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Sol | 28.28 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort high; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Terra | 16.89 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort high; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-6-Astra | 77.34 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort high; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Luna | 3.03 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort low; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Sol | 11.12 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort low; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Terra | 8.6 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort low; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-6-Astra | 59.42 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort low; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Luna | 31.02 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort max; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Sol | 49.28 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort max; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Terra | 33.46 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort max; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-6-Astra | 68.76 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort max; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.5 | 9.64 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort medium; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Luna | 8.15 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort medium; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Sol | 22.2 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort medium; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Terra | 10.1 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort medium; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-6-Astra | 68.16 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort medium; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Luna | 1.53 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort none; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Sol | 4.99 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort none; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Terra | 3.55 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort none; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Luna | 24.63 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort xhigh; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Sol | 36.24 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort xhigh; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-5.6 Terra | 25.84 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort xhigh; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GPT-6-Astra | 72.6 | points | Native editable single-slide PPTX; 500 tasks; Codex; reasoning effort xhigh; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| Kimi K3 | 67.8 | points | Native editable single-slide PPTX; 500 tasks; OpenCode; reasoning effort high; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |
| GLM-5.3-Flash | 32.42 | points | Native editable single-slide PPTX; 500 tasks; OpenCode; reasoning effort max; artifact validation; GPT-5.6 Luna High judge, 3 rounds; majority semantic/render gate; deterministic 30/40/30 detail deductions; mean over all tasks including gated zeros | 2026-09-28 · document_publication | PPTBench paper v2 · Table 1 |