Models by use case - boards verified October 17, 2026
Best AI models by use case
6 use cases, each ranked by one public benchmark board in the board's own order, with every model's API price, context window and weights from the model registry. Each page names the leader, the cheapest priced model within 5 points of the top score and the highest-placed open-weight model, and is published only while its board has entries from the last 120 days before it was checked.
Use cases
DeepSWE v1.1
Coding
- Leader
- GPT-6 Astra, Gemini 3.8 Flash and Claude Opus 5 (tied)
- Score
- 74%±3% / 74%±1% / 74%±4%
- Ranked
- 21 models
- Data thru
- Sep 22, 2026
Terminal-Bench 4.0
Agentic coding
- Leader
- Claude Opus 5.5
- Score
- 64.8% ± 3.1%
- Ranked
- 23 models
- Data thru
- Sep 29, 2026
GPQA Diamond
Reasoning
- Leader
- GPT-6 Astra
- Score
- 96.0%
- Ranked
- 20 models
- Data thru
- Oct 15, 2026
OSWorld
Computer use
- Leader
- Claude Fable 5
- Score
- 86.0%
- Ranked
- 21 models
- Data thru
- Jul 31, 2026
tau3-Banking
Tool use
- Leader
- Qwen3.8-Max
- Score
- 55.2%
- Ranked
- 17 models
- Data thru
- Aug 4, 2026
LMArena Text
Chat
- Leader
- Gemini 4 Argon
- Score
- 1524±9 (Preliminary)
- Ranked
- 22 models
- Data thru
- Oct 15, 2026
Open-weight rankings exist where a board ranks at least 3 models the registry lists with open weights. A single benchmark is not a verdict: each page states its board, its date and how settings differ between rows.
Held back until their boards refresh: Math (AIME 2026: MathArena lists AIME 2026 as deprecated and current frontier models have no entry); Vision (MMMU, newest entry Nov 24, 2025); Factual accuracy (SimpleQA, newest entry Jul 11, 2025).
Newsletter
Get new benchmark results as they land
One email per week. New model releases, the leaderboard moves that matter and what they mean for what you build.