Key findings
- Claude Fable 5.1 has the top quality score in 8 fields and Claude Opus 5.5 in the other 4.
- The value pick is a GPT model in every field: GPT-6.1 Sol in 8 fields, GPT-6 Astra in 3 and GPT-6 Sol in 1.
- In presentations, legal and medical work the second place is within 13 points of the first and costs less, so it is worth considering.
- Some leaders have run only part of their field so far, such as Claude Fable 5.1 on Claude Code with 8 of 20 data analysis tasks.
Picks for each field
Best quality is first place on the field's board. Best value is the highest-scoring pairing that costs no more than a third of the best quality pick per task.
| Field | Best quality | Score and cost | Best value | Score and cost |
|---|---|---|---|---|
| Presentations | Claude Fable 5.1 · OpenCode | 1446 · $4.41 | GPT-6 Astra · Codex | 1262 · $1.25 |
| Documents | Claude Fable 5.1 · Claude Code | 1336 · $1.37 | GPT-6.1 Sol · OpenCode | 1245 · $0.22 |
| Visual design | Claude Fable 5.1 · OpenCode | 1413 · $0.68 | GPT-6.1 Sol · OpenCode | 1320 · $0.13 |
| Data analysis | Claude Fable 5.1 · Claude Code | 1413 · $0.22 | GPT-6.1 Sol · Codex | 1283 · $0.07 |
| Website | Claude Opus 5.5 · Claude Code | 1307 · $3.41 | GPT-6.1 Sol · Codex | 1187 · $0.43 |
| 3D | Claude Opus 5.5 · Claude Code | 1327 · $17.73 | GPT-6.1 Sol · Codex | 1259 · $3.70 |
| Finance | Claude Fable 5.1 · OpenCode | 1303 · $1.28 | GPT-6.1 Sol · OpenCode | 1234 · $0.41 |
| Medical | Claude Opus 5.5 · Claude Code | 1356 · $0.57 | GPT-6 Sol · OpenCode | 1293 · $0.10 |
| Legal | Claude Fable 5.1 · Claude Code | 1402 · $4.76 | GPT-6 Astra · Codex | 1264 · $1.31 |
| Consulting | Claude Fable 5.1 · Claude Code | 1370 · $3.03 | GPT-6 Astra · Codex | 1270 · $0.76 |
| Education | Claude Opus 5.5 · Claude Code | 1353 · $0.20 | GPT-6.1 Sol · Codex | 1196 · $0.04 |
| Audio & video | Claude Fable 5.1 · OpenCode | 1467 · $0.52 | GPT-6.1 Sol · Codex | 1267 · $0.07 |
Slides documents and data
Presentations. Claude Fable 5.1 on OpenCode leads with 1446, but Claude Opus 5.5 on Claude Code is one point behind at less than half the cost, $2.12 against $4.41, and finishes runs about 30% faster. For most slide work the Opus pairing is the better buy. GPT-6 Astra on Codex is the value pick at $1.25.
Documents. Reports, document processing and news editing. Claude Fable 5.1 on Claude Code leads at 1336, but has run 16 of the 28 tasks. GPT-6.1 Sol on OpenCode scores 1245 for $0.22 a task, about a sixth of the leader's cost.
Data analysis. Spreadsheets and dashboards are cheap to run: the leader costs $0.22 a task. Claude Fable 5.1 on Claude Code leads at 1413 from 8 of 20 tasks; Claude Opus 5.5 on Claude Code has run all 20, scores 1381 and is the fastest of the top ten at 95.7 seconds a run.
Visual design. Posters and infographics. With six tasks, this board can still change. Claude Fable 5.1 on OpenCode leads at 1413, and GPT-6.1 Sol on OpenCode scores 1320 for a fifth of the cost.
Websites and 3D
Website. Claude Opus 5.5 on Claude Code leads by 65 points over the next pairing, the same model on OpenCode. The value pick, GPT-6.1 Sol on Codex, scores 1187 for $0.43.
3D. 3D assets and interactive scenes are by far the most expensive field. Claude Opus 5.5 on Claude Code leads at 1327 but spends $17.73 a task and takes 2184.2 seconds a run. GPT-6.1 Sol on Codex places fifth at 1259 for $3.70 and 959.9 seconds.
Finance medical legal and consulting
Finance. Financial research, accounting and audit, and financial customer service. Claude Fable 5.1 on OpenCode leads at 1303. The field is tight: fifth place, GPT-6.1 Sol on OpenCode, is 69 points behind at a third of the cost.
Medical. Clinical reasoning, nursing, healthcare operations and social work. Claude Opus 5.5 on Claude Code leads at 1356, with Claude Fable 5.1 on OpenCode two points behind, slightly cheaper and faster. GPT-6 Sol on OpenCode is the value pick at $0.10 a task.
Legal. With 93 tasks across eight practice areas, the largest field on the board. Claude Fable 5.1 on Claude Code leads at 1402 from 55 tasks. Claude Opus 5.5 on Claude Code has run all 93, scores 1389 and costs $3.42 against $4.76.
Consulting. Tax, operations, procurement and administrative work. Claude Fable 5.1 on Claude Code leads at 1370, but from 20 of the 56 tasks. The next three pairings, all within 90 points, have run more than 50 each. GPT-6 Astra on Codex scores 1270 for $0.76.
Education and audio
Education. Knowledge assessment and roleplay tasks. Claude Opus 5.5 on Claude Code leads at 1353 for $0.20 a task, 64 points clear of second place. GPT-6.1 Sol on Codex costs $0.04.
Audio & video. Six tasks so far, so read the order as provisional. Claude Fable 5.1 on OpenCode leads at 1467, and GPT-6.1 Sol on Codex costs $0.07 a task.
Before you choose
- Field scores cannot be compared with each other. A 1400 in legal and a 1300 in finance say nothing about which field the pairing is better at; compare places instead.
- Where the leader has run only part of the field, the second or third pairing may have a better-supported score. The table above flags the largest cases.
- Cost is model spend only. A pairing that needs more human review can cost more in practice than its token bill suggests.
Each field page also has subcategory boards, such as accounting and audit within finance or litigation within legal. The quality and cost analysis explains how the value picks compare on the overall board.
Questions
What is the best AI agent for making presentations
On Work Arena's presentations board, Claude Fable 5.1 on OpenCode scores 1446 and Claude Opus 5.5 on Claude Code 1445. The Opus pairing costs $2.12 a task against $4.41, so it is the better choice for most slide work. GPT-6 Astra on Codex scores 1262 for $1.25.
What is the best AI agent for legal work
Claude Fable 5.1 on Claude Code leads the legal board at 1402, and Claude Opus 5.5 on Claude Code is second at 1389 after running all 93 legal tasks. GPT-6 Astra on Codex is the strongest option at under a third of the cost, at 1264.
What is the best AI agent for data analysis and spreadsheets
Claude Fable 5.1 on Claude Code leads at 1413 and Claude Opus 5.5 on Claude Code scores 1381, both for about $0.21 a task. GPT-6.1 Sol on Codex scores 1283 for $0.07.
Figures in this article come from the leaderboard data of 2 November 2026. The leaderboard shows any changes since.