ENC-Bench - Track Direction Recognition: leaderboard

Metric: Accuracy (%) on ENC-Bench's 141 track direction recognition questions (Decision-Making: giving the legal navigation direction of a marked route, either way or non-navigable; three options), on ENC-Bench's 840 NOAA S-57 electronic navigational charts rendered with OpenCPN in day, dusk and night palettes at six scales, zero-shot; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 235B A22B (Thinking)75.18#228 (Qwen 3 VL 235B A22B)
2Qwen 3 VL 235B A22B Instruct74.47#264
3Qwen 3 VL 32B (Thinking)73.76#287 (Qwen 3 VL 32B)
4Qwen 3 VL 32B Instruct68.79#276
5Gemini 2.5 Pro63.12#145
6Gemini 2.5 Flash59.57#237
7Llama 4 Maverick Instruct53.9#439
8GLM-4.5V53.19#339
9InternVL3-38B51.06#395
10GPT-4o45.39#333

No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.

Interactive version: theaggregate.ai/benchmark?slug=enc-bench-track-direction-recognition · How It Works · Data refreshed daily, snapshot 2026-10-11.