IBM logo

TTM-R3

online
ibm-research/ttm-r3

~1.4M (Lite) to ~35M params | 512 context | $0.5000 input | $1.50 output | CC-BY-NC-SA-4.0

TTM-R3 is IBM Research's March 31, 2026 refresh of the TinyTimeMixer family. While still classified as `tinytimemixer` under the hood, R3 adds trend-residual decomposition, a multi-quantile probabilistic forecasting head, gated attention, FFT-based frequency embeddings, and learnable sequence-level register tokens, while preserving the compact ~1.4M–35M parameter footprint that makes TTM practical on CPUs and lightweight hosted inference. IBM reports a 15–50x inference speedup vs state-of-the-art forecasters and a meaningful accuracy improvement over TTM-R2. Hosted on TSFM.ai under a pass-through compute posture: TTM-R3 ships under the CC-BY-NC-SA-4.0 license (research / non-commercial use only), so you are responsible for ensuring your intended use falls within the upstream license — see section 7 of our Terms of Service.

Model Classification

Family

TinyTimeMixer

Type

time series foundation model

Pretrained time-series model exposed on TSFM.ai for zero-shot or few-shot forecasting workloads.

Training Data

IBM's GiftEvalPretrain subset plus KernelSynth-style synthetic augmentation; corpus aligned with the TTM family rather than narrowed to a single benchmark.

Recommended For

  • CPU-friendly or latency-sensitive forecasting baselines
  • Fast zero-shot checks before escalating to larger TSFMs

Strengths

  • Very small checkpoints with efficient deployment characteristics
  • Useful lightweight baseline for standard public forecasting workloads

Limitations

  • Lower ceiling than larger modern TSFM families on broad zero-shot leaderboards
  • Checkpoint families are tuned around specific context and prediction settings

Capabilities

forecastingquantile-forecastingprobabilistic-forecastingmultivariatezero-shothigh-throughput

Tags

ibmibm-researchttmtinyresearch-licensenon-commercialprobabilistic

Specifications

Parameters
~1.4M (Lite) to ~35M
Architecture
TinyTimeMixer with trend-residual decomposition, gated attention, multi-quantile head, and FFT embeddings
Context length
512
Max output
1,024
Avg latency
n/a
Uptime
n/a
Plan limits
1,000 rpm free · 1,000,000 rpm with billing
Accelerator
NVIDIA GPU
Regions
Virginia, US
License
CC-BY-NC-SA-4.0

Pricing

Input / 1M tokens
$0.5000
Output / 1M tokens
$1.50

Performance

Average latency
n/a
Availability
n/a
Plan limits
1,000 rpm free · 1,000,000 rpm with billing