Treble Technologies and Hugging Face have jointly launched the Far-Field Automatic Speech Recognition Leaderboard, the first open, community-driven benchmark designed to evaluate voice models under realistic acoustic conditions. Available now on Hugging Face Spaces, the platform addresses a persistent industry challenge: the significant performance degradation that occurs when near-field models are deployed in reverberant, noisy environments typical of smart assistants, in-car systems, and robotic voice interfaces. Traditional evaluation datasets rely on clean, close-microphone recordings that fail to simulate real-world acoustic complexity. The FFASR benchmark corrects this by utilizing Treble Technologies proprietary hybrid simulation engine, which combines wave-based solvers with geometrical-acoustics modeling to generate highly realistic audio. The initial test suite encompasses 14 fully furnished rooms ranging from 20 to 470 cubic meters, including offices, classrooms, and domestic spaces. Each scenario features a primary speaker alongside multiple transient and continuous noise sources, evaluated across three signal-to-noise ratio tiers. A beta track for moving-source audio further simulates dynamic environments such as mobile devices and walking robots. The leaderboard tracks two primary metrics: Word Error Rate and Real-Time Factor, measured on standardized NVIDIA L4 hardware. This dual-focus approach allows developers to visualize accuracy versus inference speed using Pareto front analysis, revealing trade-offs that clean-speech benchmarks obscure. Early submissions demonstrate a stark divergence between near-field and far-field performance, with word error rates at low signal-to-noise ratios frequently exceeding near-field baselines by multiple factors. The platform intentionally reports both metrics side-by-side, enabling engineers to distinguish between models with inherent recognition accuracy and those requiring robust acoustic conditioning or preprocessing pipelines. Submissions operate server-side via the Hugging Face Hub, supporting mainstream architectures including Whisper variants, IBM Granite, Wav2Vec2, HuBERT, and SpeechBrain. Custom inference stacks with integrated speech enhancement can be deployed through moderated custom evaluator jobs. The evaluation set comprises 2,000 held-out anechoic samples across all conditions, ensuring consistent normalization and preventing test contamination. Treble and Hugging Face have outlined a clear roadmap for the benchmark, with upcoming tracks targeting multi-talker scenarios, microphone array beamforming, and active echo cancellation. The initiative aims to shift research priorities toward real-world acoustic robustness, providing standardized visibility into model performance where it matters most. Developers and researchers are invited to submit models, analyze deployment-specific trade-offs, and contribute to the benchmark evolution through the dedicated FFASR forum.