ts-bench v1: leaderboard

A command-line coding agent driving the named model edits 25 Exercism TypeScript exercises until their test suites pass; the score is the share of the 25 solved in one run per agent and model. Aider Polyglot uses Exercism too, across six languages and with one fixed harness.

Metric: Success rate (%). Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 15 models tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-596#177
2Gemini 3 Pro (Preview)96#120
3GPT-5.3 CodexHigh96#122
4Gemini 2.5 Pro92#274
5Gemini 3 Flash (Preview)92#151
6GLM-4.692#496
7GPT-5.2High92#185
8Qwen3 Coder Next92#724
9Claude Sonnet 4.588#258
10GPT-5.1 CodexHigh88#284
11Gemini 2.5 Flash (Preview 09-2025)84#449
12GLM-4.580#537
13Qwen 3 Coder Plus (2025-09-23)76#708
14Claude Sonnet 4 (20250514)72#405
15Qwen 3 Coder Plus (2025-07-22)64

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=ts-bench-v1 · How It Works · Data refreshed daily, snapshot 2026-11-17.