What newer quantization options - beyond the original scalar, product, and binary trio - has Qdrant introduced to fine-tune the compression/accuracy curve?

Qdrant has expanded its quantization menu beyond the original scalar/product/binary trio in recent releases, primarily to fill the gap between int8 (4x) and binary (32x), and to reduce the accuracy cost of binary without giving up its memory savings. The most significant additions are sub-byte quantization - schemes that use fewer than 8 bits per dimension but more than 1, such as 1.5-bit and 2-bit variants - and asymmetric quantization, where the query vector is kept in full precision while the stored vectors are quantized. Sub-byte schemes let you pick a point on the compression curve between scalar and binary, which matters because the jump from 4x to 32x is very large and most datasets do not need to go all the way. Asymmetric quantization changes the error model: by keeping the query in float32, the distance computation is asymmetric (one operand full precision, one quantized), which can substantially reduce the quantization error compared to quantizing both sides, at the cost of a slightly slower distance computation.

The mechanism for sub-byte quantization is essentially the same as scalar quantization - a linear map from float to a small integer grid - but the grid is coarser (4 or 16 levels instead of 256). The trade-off is more accuracy loss than int8 but less than binary, with memory savings in between. Asymmetric quantization is a different idea: instead of approximating both query and stored vector with the same low-precision grid, you approximate only the stored side and compute the distance using the full-precision query. This reduces the error substantially because the query is often the dominant source of error in symmetric quantization - the query is a single vector used across many comparisons, so its quantization error is systematic rather than averaging out. By keeping the query exact, you remove that systematic error. The cost is that the distance computation is no longer symmetric, so it cannot use the same optimized SIMD kernels as symmetric quantization, and it may be slower per comparison.

The trade-off with these newer options is complexity and version dependence. The original trio is well understood and widely documented; the newer schemes are more sensitive to the specific version of Qdrant you are running and to the data distribution. My advice is to treat the newer schemes as advanced tuning knobs to reach for only after you have exhausted the basics: measure SQ first, then BQ with oversampling and rescoring, and only then look at sub-byte or asymmetric schemes if you need a point on the curve that neither SQ nor BQ can hit. The common mistake is jumping to the newest option because it sounds better, without benchmarking it against SQ on your own data. On many datasets, SQ at 4x with a slightly higher m outperforms a more exotic scheme at 8x with a lower m. The other mistake is assuming these options are universally available; they are version-dependent and sometimes limited to specific distance metrics or vector types. Version note: this is the area of Qdrant that has changed the most in recent releases, so any specific claim about sub-byte or asymmetric quantization should be verified against the release notes and the actual schema of your deployed version before you rely on it.

Version-dependent: this is the single most version-sensitive area of Qdrant covered in this bank. Sub-byte schemes, asymmetric quantization, and the exact set of configurable fields have all changed across recent releases. Do not assume a scheme is available or behaves a certain way based on a blog post or an older version of the docs. Read the release notes for your specific version, check the collection config at runtime, and benchmark on your own data before committing to a scheme in production.