# Amazon Chronos-Bolt vs Google TimesFM 2.0: 2026 Empirical Benchmark & Cost Shootout

> **Direct Answer (October 2026 Verdict)**: Amazon Chronos-Bolt wins on native probabilistic quantile accuracy (Weighted Quantile Loss on FEV-bench/GiftEval), 250x faster direct multi-step patch decoding (1.8ms/series), and tiny VRAM footprint across 6.1k+ GitHub stars (97.9), while Google TimesFM 2.0 wins for long-context (2,048-step) point and covariate-aware forecasting where external regressors (price, promotions, weather) drive demand across 34.6k+ GitHub stars (97.7).

- **Category**: Patch-Based Probabilistic T5 Time-Series Foundation Model vs Decoder-Only Covariate Foundation Model Shootout
- **Canonical URL**: https://exactbaseline.com/vs/chronos-vs-timesfm
- **1-Click $49 Builder Repo Hardening & Launch Kit (Stripe Checkout)**: https://exactbaseline.com/buy/instant_boost
- **1-Click $149 Category Leader Pro + 30-Day Pin (Stripe Checkout)**: https://exactbaseline.com/buy/category_leader

## Amazon Chronos-Bolt Key Advantages
- 6,126+ GitHub stars under Apache-2.0 by Amazon Science, achieving #1 zero-shot Weighted Quantile Loss (WQL) and MASE across 27 benchmark datasets
- Chronos-Bolt replaces autoregressive token generation with direct multi-step patch forecasting, running 250x faster and using 20x less GPU memory
- Outputs calibrated probabilistic quantile trajectories (P10, P50, P90) out of the box for inventory safety-stock and SLA capacity planning

## Google TimesFM 2.0 Key Advantages
- 34,572+ GitHub stars under Apache-2.0 by Google Research, pre-trained on 100+ billion real-world time points from Google Trends and Wiki Pageviews
- First-class support for dynamic numerical/categorical covariates and static metadata via lightweight in-context linear/MLP adapters
- Supports context lengths up to 2,048 historical time steps and arbitrary forecast horizons in PyTorch and JAX/Flax