Task Shortlist · Agents & automation

Best API models for agents & automation

Start with a cheaper model and move up when it fails.

Trade-offs & Editorial Note

Terra is the low-price visual and computer-use option among mid-tier models. Haiku 5.5 is the budget browser-use option. Complex agents may need stronger models, supervision and actions you can undo.

1

Low-cost screen work, visual reasoning and drafting.

Benchmark $0.96/task · $1.00 in / $5.00 out per 1M

51.3INDEX
Dossier
2

Classification, extraction, routing, summaries, subagents and browser use at very low cost.

Benchmark $0.41/task · $0.10 in / $0.50 out per 1M

36.5INDEX
Dossier
3
GPT-6.2 SolOpenAINew

Everyday coding, general work and tool-using agents.

Benchmark $1.01/task · $2.00 in / $10.00 out per 1M

57.3INDEX
Dossier
4

Everyday coding, multimodal work and agents.

Benchmark $3.16/task · $1.50 in / $7.50 out per 1M

53.9INDEX
Dossier
5
GPT-6.2 AstraOpenAINew

Difficult software, computer-use and document workflows.

Benchmark $4.00/task · $8.00 in / $40.00 out per 1M

62.1INDEX
Dossier