Key findings
- Claude Opus 5.5 scores higher in all 12 fields and overall, where it leads 1335 to 1235.
- The gap ranges from 26 points in finance to 159 in medical. Under 45 points in four fields, over 100 in six.
- GPT-6 Astra costs $1.28 a task overall against $2.60, and is cheaper in 11 of the 12 fields. Presentations is the exception.
- Both run best on their vendor's harness: Claude Opus 5.5 on Claude Code and GPT-6 Astra on Codex.
Overall
| Claude Opus 5.5 | GPT-6 Astra | |
|---|---|---|
| Best harness | Claude Code | Codex |
| Overall place | 1 | 5 |
| Quality | 1335 | 1235 |
| Time per run | 746.5s | 722.8s |
| Cost per task | $2.60 | $1.28 |
| Tasks run | 402 of 406 | 392 of 406 |
| On the other harness | OpenCode 1298 | OpenCode 1208 |
A 100-point gap means Claude Opus 5.5 is expected to win about 64% of same-task comparisons against GPT-6 Astra. The two take almost the same time per run. GPT-6 Astra costs about half as much.
Both models do best on their own vendor's harness. Claude Opus 5.5 loses 37 points on OpenCode and GPT-6 Astra loses 27, which is why the harness matters when comparing them.
Field by field
For each field, the table takes the best-placed pairing of each model.
| Field | Claude Opus 5.5 | GPT-6 Astra | Gap | Opus expected win rate |
|---|---|---|---|---|
| Medical | 1356 · $0.57 | 1197 · $0.53 | 159 | 71% |
| Presentations | 1445 · $2.12 | 1300 · $2.33 | 145 | 70% |
| Education | 1353 · $0.20 | 1210 · $0.17 | 143 | 69% |
| Legal | 1389 · $3.42 | 1270 · $2.39 | 119 | 66% |
| Audio & video | 1363 · $0.56 | 1253 · $0.21 | 110 | 65% |
| Website | 1307 · $3.41 | 1202 · $1.18 | 105 | 65% |
| 3D | 1327 · $17.73 | 1245 · $8.76 | 82 | 62% |
| Visual design | 1398 · $0.85 | 1322 · $0.29 | 76 | 61% |
| Documents | 1297 · $0.90 | 1253 · $0.87 | 44 | 56% |
| Data analysis | 1381 · $0.21 | 1344 · $0.17 | 37 | 55% |
| Consulting | 1303 · $2.61 | 1270 · $0.76 | 33 | 55% |
| Finance | 1256 · $1.45 | 1230 · $0.53 | 26 | 54% |
When Claude Opus 5.5 is the better choice
- Presentations. The gap is 145 points and GPT-6 Astra is not cheaper here: its best pairing costs $2.33 a task and takes nearly twice as long, 1909.1 seconds against 976.5.
- Medical and education. Gaps of 159 and 143 points, while both models cost under $0.60 a task. With prices this close, there is little reason to take the weaker result.
- Data analysis and documents. The gaps are smaller, 37 and 44 points, but so is the price difference: a few cents a task.
- Legal and websites. Gaps above 100 points. GPT-6 Astra saves about $1 to $2 a task, which may matter at volume, but the quality difference is large.
When GPT-6 Astra is the better choice
- Finance. The gap is 26 points, an expected win rate of 54% for Claude Opus 5.5, and GPT-6 Astra on Codex costs $0.53 a task against $1.45.
- Consulting. A 33-point gap, and GPT-6 Astra on Codex costs $0.76 against $2.61 while finishing in 517.4 seconds instead of 1165.7.
- High-volume work in any field. Across the overall board GPT-6 Astra delivers 1235 for half the cost, which still places fifth of 49.
A cheaper OpenAI option
Within OpenAI's line-up, GPT-6.1 Sol is often the better buy than GPT-6 Astra. Overall it scores 1196 on OpenCode for $0.51 a task, 39 points below GPT-6 Astra on Codex at 40% of the cost. In four fields it scores higher than GPT-6 Astra outright:
| Field | GPT-6.1 Sol | GPT-6 Astra |
|---|---|---|
| Medical | 1292 · $0.14 | 1197 · $0.53 |
| Audio & video | 1267 · $0.07 | 1253 · $0.21 |
| 3D | 1259 · $3.70 | 1245 · $8.76 |
| Finance | 1234 · $0.41 | 1230 · $0.53 |
The quality and cost analysis shows where GPT-6.1 Sol sits on the overall frontier.
How this comparison was made
Every figure comes from the Work Arena leaderboard. Both models ran the same tasks and were graded against the same criteria. Quality is fitted from wins and losses on the same task, so it measures how often one pairing's work beats another's, not a percentage of tasks passed. Cost is model spend at official prices. The methodology has the full rules. Claude Opus 5.5 ran on Claude Code and OpenCode, GPT-6 Astra on Codex and OpenCode.
Questions
Is Claude Opus 5.5 better than GPT-6 Astra
On Work Arena's real work tasks, yes in every field. Claude Opus 5.5 on Claude Code scores 1335 overall against 1235 for GPT-6 Astra on Codex. The gap is small in finance, consulting, data analysis and documents and large in medical, presentations, education and legal.
Is GPT-6 Astra cheaper than Claude Opus 5.5
Yes, in 11 of 12 fields. Overall GPT-6 Astra on Codex costs $1.28 a task against $2.60 for Claude Opus 5.5 on Claude Code. In presentations GPT-6 Astra is slightly more expensive, $2.33 against $2.12.
Figures in this article come from the leaderboard data of 2 November 2026. The leaderboard shows any changes since.