Resyst Labs Benchmarks
← Back to overview

Resyst Arena · replay room

Tactical evidence you can replay.

Resyst Arena is a deterministic turn-based evaluation environment for spatial strategy, legal-action discipline, and long-horizon tactical continuity. Replays are grouped by encounter so historical DeepSeek vs Step, DeepSeek vs Gemini, and DeepSeek vs Kimi runs stay readable instead of collapsing into one flat match list.

Browse encounters Download Arena data
Score boundaryOutcome first

Latency, token use, and cost are telemetry, not hidden score modifiers.

Encounter groupingPairing first

Replay buttons live under the model-vs-model encounter they belong to, including side-swapped rounds.

Replay contractSanitized state

Replay JSON exposes board states, actions, events, and telemetry while excluding raw model text outputs.

Encounter 1

DeepSeek V4 Flash vs Kimi K2.7 Code

Kimi K2.7 Code leads 2–0

Replays2
Turns120
Seeds1

2 replays · latest 20260612 · extra-high reasoning text actions

Encounter 2

DeepSeek V4 Flash vs Step 3.7 Flash

DeepSeek V4 Flash leads 2–0

Replays2
Turns163
Seeds2

2 replays · latest 20260531 · 0.3

Encounter 3

DeepSeek V4 Flash vs Gemini 3 Flash Preview

Gemini 3 Flash Preview won the replay

Replays1
Turns200
Seeds1

1 replay · latest 20260530 · 0.3

Seed 361605864 · duel · 60 turns

DeepSeek V4 Flash vs Kimi K2.7 Code

A: DeepSeek V4 Flash B: Kimi K2.7 Code

A
DeepSeek V4 Flash Side A
Core integrity 27 / 30
Energy reserve — / 12
Damage1
Resources23
Invalid0
Units—
B
Kimi K2.7 Code Side B
Core integrity 29 / 30
Energy reserve — / 12
Damage4
Resources17
Invalid0
Units—
Loading model Board state · legal actions · event telemetry
Replay control
Loading replay…
A core B core Control zone Resource Current action