Overview
Scribe v2 is ElevenLabs' batch speech-to-text model for long-form audio, released January 12, 2026 per the official changelog. Designed for high-accuracy transcription of complex content like podcasts, interviews, and media files, this version introduces enterprise-grade features including keyterm prompting, entity detection, and comprehensive compliance controls. With support for over 90 languages and automatic multilingual detection, Scribe v2 targets professional workflows requiring precision transcription at scale—from content production and research to legal and medical documentation.
What's New
Accuracy Improvements
ElevenLabs states that Scribe v2 achieves the lowest word error rates on industry-standard benchmarks. The official Speech-to-Text page presents benchmark comparisons with GPT-4o Transcribe, Gemini 2.5 Pro, and AssemblyAI, showing Scribe v2's competitive performance. The model demonstrates significantly improved robustness in handling long-form audio, tone changes, extended silences, and pauses—challenging conditions that previously degraded transcription quality in earlier versions.
Keyterm Prompting
Users can now provide up to 100 domain-specific terms—such as brand names, technical jargon, product identifiers, or proper nouns—to guide the model's transcription. This feature ensures accurate rendering of specialized vocabulary that standard language models might misinterpret, particularly valuable for technical documentation, legal proceedings, medical transcription, and industry-specific content where terminology precision is critical.
Entity Detection with Timestamps
Scribe v2 natively detects and tags up to 56 categories of entities, including personally identifiable information (PII), health data, payment details, locations, and organization names. Each detected entity is returned with timestamps, enabling automated review workflows, compliance auditing, and selective redaction for sensitive content. This capability streamlines production pipelines that require entity extraction for indexing, privacy compliance, or content moderation.
Multilingual Audio Processing
The model automatically detects and transcribes multilingual audio within a single file, eliminating the need for manual segmentation when speakers switch languages mid-conversation. This feature supports international meetings, cross-border interviews, and content with code-switching patterns, reducing preprocessing time and improving workflow efficiency for global media operations.
Dynamic Audio Tagging
Scribe v2 automatically identifies and labels non-speech audio events such as laughter, applause, and footsteps. These tags appear in the transcript with timestamps, providing richer context for post-production teams working on subtitles, content analysis, or quality assurance workflows where audio context matters beyond spoken words.
Enterprise Compliance and Data Controls
This release introduces comprehensive compliance controls aligned with SOC 2, ISO 27001, PCI DSS Level 1, HIPAA, and GDPR requirements. Organizations can select EU or India data residency options to ensure audio processing remains within specific geographic boundaries. A zero-retention mode allows immediate deletion of audio files after transcription completes, meeting strict data handling policies for sensitive use cases in healthcare, legal, and financial services.
Pricing & Plans
ElevenLabs plans include a monthly allowance of transcription hours for Speech-to-Text. Additional transcription beyond the included hours is billed per hour (the pricing page lists rates like $0.40/hour). Advanced features including entity detection and keyterm prompting are offered as paid add-ons. All plans also include access to text-to-speech, music generation, and AI agent features with separate credit allocations.
- Free — $0/month, includes transcription hours (check pricing page for current limits), 3 Studio projects, API access
- Starter — $5/month, includes additional transcription hours, commercial license, 20 Studio projects
- Creator — $22/month, includes expanded transcription hours, first-month promotional pricing available on the pricing page
- Pro — $99/month, higher transcription hour allowance
- Scale — $330/month, increased transcription capacity, 3 workspace seats
- Business — $1,320/month, enterprise-level transcription volume, 5 workspace seats
- Enterprise — Custom pricing with custom transcription volumes, SLAs, security features (HIPAA, SSO), priority support
For paid subscriptions, unused credits can roll over for up to two months according to ElevenLabs billing documentation. Specific transcription hour allocations and add-on pricing are detailed on the official pricing page.
Best For
- Podcast producers and media companies requiring accurate long-form transcription with minimal post-production editing for subtitles and show notes
- Research teams conducting multilingual interviews where automatic language detection eliminates manual segmentation overhead
- Legal and healthcare organizations needing HIPAA and GDPR-compliant transcription with entity detection for PII and sensitive data handling
- Technical content creators working with specialized vocabulary who benefit from keyterm prompting for accurate rendering of domain terminology
- Enterprise media operations requiring speaker diarization for up to 48 participants with word-level timestamps for subtitle synchronization
- Global businesses processing international meetings with code-switching patterns across 90+ supported languages
FAQ
How does Scribe v2 compare to Scribe v1 in accuracy?
ElevenLabs states that Scribe v2 achieves the lowest word error rates on industry benchmarks, representing a significant improvement over v1. Key enhancements include better handling of tone changes, extended silences, pauses, and long-form audio—challenging conditions that previously degraded transcription quality. The official product page presents benchmark comparisons showing Scribe v2's performance against current competing models.
Can I use Scribe v2 for real-time transcription?
Scribe v2 is optimized for batch processing of pre-recorded audio files. For real-time transcription, ElevenLabs offers Scribe v2 Realtime as a separate model variant. The Realtime product page describes latency under 150ms, while specific integration contexts (such as Agents) mention latency ranges of 30-80ms. Both models share similar accuracy characteristics, but Realtime is specifically engineered for live speech input, voice agents, and streaming applications where immediate response is required.
What languages does keyterm prompting support?
Keyterm prompting allows you to provide up to 100 domain-specific terms or phrases per transcription request to bias the model toward specific vocabulary. This feature helps improve transcription accuracy for specialized terminology, proper nouns, brand names, and technical jargon that standard language models might misinterpret. The feature is available for Scribe v2 transcriptions across its supported languages.
Do I need an Enterprise plan to use compliance features?
ElevenLabs states support for SOC 2, HIPAA, GDPR compliance and offers EU data residency and zero-retention modes. However, specific plan availability and contractual requirements (such as Business Associate Agreements for HIPAA) may vary by tier and use case. Check your workspace settings or contact ElevenLabs sales to confirm which compliance features and data controls are included in your plan and whether additional agreements are required for your regulatory needs.
How is Speech-to-Text usage billed?
Each ElevenLabs plan includes a monthly allowance of transcription hours. Additional transcription beyond the included hours is billed per hour (for example, the pricing page lists rates like $0.40/hour for additional usage). Advanced features such as entity detection and keyterm prompting are billed as separate add-ons. Note that Speech-to-Text billing is separate from the credit system used for text-to-speech and other platform features. Review the official pricing page for current transcription hour allocations by plan tier.
