• 5 min read

Releasing Project Solace 1.0 Omni: 12.59M Conversations Across 60 Datasets

By Solstice-AI Core Team
KEY TAKEAWAYS
  • Solace 1.0 Omni aggregates 12,586,893 unique conversations across 60 audited source datasets.
  • Exact SHA256 deduplication removed 3,445,534 redundant copies (21.5% overlap).
  • Released in drop-in OpenAI messages format under AGPL-3.0 on Hugging Face (34.3 GB compressed).

Today we are releasing Project Solace 1.0 Omni, the largest verified frontier-model distillation corpus ever released — comprising 12,586,893 unique conversations (137.7 GB uncompressed / 34.3 GB compressed) in drop-in OpenAI messages format.

Why Multi-Teacher Distillation?

Most distillation datasets rely on a single teacher model. This creates an echo-chamber effect where the student model inherits the teacher’s idiosyncratic weaknesses, phrasing tropes, and blind spots.

Solace 1.0 Omni samples problem-solving paths from 7 verified frontier model families:

  • GLM 5.2 (15 datasets): OpenHands agent rollouts, on-policy regeneration, and ARC traces.
  • Claude Fable 5 & Mythos 5: Anthropic’s reasoning engines and agentic coding sessions.
  • GPT-5.6 Sol & GPT-5.5 Codex: Coding, debugging, and ARC-AGI3 challenges.
  • DeepSeek V4 Pro 0813: 200K math/STEM and SWE-bench replays.
  • Qwen 3.8-Max: Cross-architecture distillation data.
  • Kimi K3 / Opus 4.7 Multi: Long-context synthesis and reasoning specialists.
  • Manus Agents: Tool use and browser execution transcripts.

Key Corpus Metrics (Measured, Not Estimated)

Metric Value
Unique Conversations 12,586,893
Source Rows Ingested 16,032,427
Exact Duplicates Purged 3,445,534 (21.5% cross-dataset overlap)
Uncompressed Size 137.7 GB
Compressed File solace_final.jsonl.gz — 34.3 GB
Source Datasets 60 Audited Datasets
Ordering Weighted round-robin interleaved

How to Access the Data

The complete dataset is hosted on Hugging Face:

from datasets import load_dataset

dataset = load_dataset(
    "Solstice-AI/Solace-1.0-Omni",
    data_files="solace_final.jsonl.gz",
    split="train"
)
print(f"Total samples: {len(dataset):,}")

For technical methodology and dedup details, read the Solace 1.0 Technical Report.