Inkling 2 Small Launch - Internal Code Bits Per Byte: leaderboard
Publisher primary held-out code evaluation, cutoff June 2026. Base-recovered models were fine-tuned on pretrain-formatted code and documents before loss evaluation. Preserve base-recovered, pre-trained and post-trained states as different systems. No public weights/release date for experimental variants is established.
Metric: Bits per byte. Source: thinkingmachines.ai. Saturation forecast: Rough model projection: around 2026. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Inkling 2 Small pre-trained (Benchmark Sweep Run: source=Inkling2SmallInternalCode; cutoff=2026-06; checkpoint state as published) | 0.2 | |
| 2 | Kimi K3.1 Pro base-recovered (Benchmark Sweep Run: source=Inkling2SmallInternalCode; cutoff=2026-06; checkpoint state as published) | 0.21 | |
| 3 | Inkling 2 Small post-trained (Benchmark Sweep Run: source=Inkling2SmallInternalCode; cutoff=2026-06; checkpoint state as published) | 0.21 | |
| 4 | DeepSeek V4.1 Flash 1017 base-recovered (Benchmark Sweep Run: source=Inkling2SmallInternalCode; cutoff=2026-06; checkpoint state as published) | 0.22 | |
| 5 | GLM-5.3 Flash base-recovered (Benchmark Sweep Run: source=Inkling2SmallInternalCode; cutoff=2026-06; checkpoint state as published) | 0.23 | |
| 6 | DeepSeek V4 Pro Base (Benchmark Sweep Run: source=Inkling2SmallInternalCode; cutoff=2026-06; checkpoint state as published) | 0.23 | |
| 7 | GLM-5.3 base-recovered (Benchmark Sweep Run: source=Inkling2SmallInternalCode; cutoff=2026-06; checkpoint state as published) | 0.24 |
Interactive version: theaggregate.ai/benchmark?slug=inkling-2-small-launch-internal-code-bits-per-byte · How It Works · Data refreshed daily, snapshot 2026-11-06.