ElevenLabs

Last verified: Jun 7, 2026

16 sourced changes·94% official·verified against official docs, changelogs & Wayback·last checked Jun 7, 2026

01 / VERDICT

Caution

Frequent recent tightening — high constraint volatility right now.

WHY CAUTION — trend: unstable · drift -70% over 90d · 7 tightenings in 90d

Constraint Stability

28/100

higher = fewer recent restrictions

unstable-70% / 90d

ElevenLabs has shown frequent recent tightening, mainly in model availability.

Most noticeable: Removes existing audio filtering capabilities from ElevenAgents configuration, potentially breaking workflows relying on these filters.

Recency

days since it last tightened

Velocity

6.3

tightenings per year

Silent

of tightenings were unannounced

Reversals

of tightenings later reversed (same dimension)

ElevenLabs constraint metrics — computed from 8 tracked tightenings · median 7 days between tightenings. Original to this site; not published by the vendor.

IMPACT BY PERSONA

Solo / occasional

Brief, occasional use is rarely touched by limit changes.

low

Power user / daily heavy

Exposed to usage-limit tightening.

medium

Team / shared workflow

Exposed to quota, model-availability changes across the team.

high

Startup / cost-scaling

Pricing and usage-based cost have been predictable as volume grows.

low

Enterprise / long-term dependency

Continuity exposure: model-continuity.

high

02 / ANALYSIS

Deprecation, not rate or price squeezes, is the through-line on ElevenLabs' developer surface. The eleven_monolingual_v1 and eleven_multilingual_v1 models were marked for removal in late 2025, and a June 2026 notice set a firm July 9, 2026 removal date for those v1 models alongside scribe_v1 — recurring forced migrations that break any integration pinned to a specific model. A user report captures the cost precisely: pinning eleven_multilingual_v1 in production, then having to re-validate every voice once a removal date appeared. Pricing has generally eased access — Eleven v3 launched with a temporary 80% discount in mid-2025 and reached general availability in February 2026 — and several capacity limits have risen, with agent-test batches expanded from 1,000 to 5,000 and SIP UUI data length lifted from 256 to 8,192 characters. Not everything loosens, though: a new zero-retention mode for text-to-dialogue was gated to enterprise customers only, and an InteractionBudget change dropped the async option for fixed windows, requiring a one-time config migration. Quantitative limits trend upward while the model catalog turns over on published schedules. Generous on capacity and reasonable on price, the platform nonetheless demands active maintenance: track the changelog, avoid hard-pinning deprecated models, and expect that privacy or newer features may sit behind the enterprise tier.

03 / CURRENT LIMITS

Current state + recent changes

Free: Limited monthly credits; non-commercial
Starter / Creator: Character credits/mo; standard TTS = 1 credit/char, Turbo/Flash = 0.5/char
Pro / Scale / Business: Higher monthly credit pools + usage-based overage

Last verified: Jun 7, 2026

RECENT LIMIT & PRICING CHANGES

BeforeAfter

Widget file-input configuration had no in-memory file cap↓Widget file-input configuration adds max_files_in_memory (integer, 1–30, default 10 ). Files beyond the limit are summarized and released from memory

Aug 2026·source

Get WebRTC token responses returned token only↓Get WebRTC token responses now require conversation_id (string) in addition to token. This is a breaking schema change for clients that validate or deserialize the response shape.

Jul 2026·source

Backup LLM cascade_timeout_seconds default: 8 seconds↓Backup LLM cascade_timeout_seconds default: 4 seconds (allowed range remains 2–15s)

Jul 2026·source

N/A→Added optional usage_limit for monthly invitee credit cap

Jul 2026·source

04 / WHERE IT MATTERS

Which usage patterns the recent changes affect — each line backed by the events driving it.

Specific-model dependencehigh exposure

Aug 2026↓Removes existing audio filtering capabilities from ElevenAgents configuration, potentially breaking workflows relying on these filters.·source

Jul 2026↓Breaking change for API clients requiring schema updates to include conversation_id in WebRTC token responses.·source

Jul 2026↓Shorter default fallback window for the backup LLM cascade in ConvAI agents; remains user-configurable within the 2–15s range.·source

Jul 2026↓Removes the ability to manually optimize streaming latency for TTS agents.·source

Jun 2026↓Users must migrate to v2 models before July 9, 2026, or lose access to these specific models.·source

Jun 2026↓Users must migrate to scribe_v2 or scribe_v2_realtime before July 9, 2026.·source

Sep 2025↓Two legacy TTS models removed; pinned integrations break·source

Long-context workloadsmedium exposure

Aug 2026↓Limits the number of files held in memory for widget file-input, requiring summarization for larger sets.·source

Sustained sessionslow exposure

Jun 2026→Dev-facing enum change: async option replaced by fixed time windows; existing configs need a one-time migration.·source

05 / WHAT CHANGED

Technical 7 · Product 7 · Business 1 · Policy 1

Feb 2026
→ v3 promoted to GA
productImpact Low · source
Sep 2025
↓ Two legacy TTS models removed; pinned integrations break
productImpact Medium · source
Jun 2025
↑ v3 launched with a temporary 80% discount (expired end of June)
businessImpact Low · source

▸ View full constraint history (16 events)

Aug 2026
↓ Limits the number of files held in memory for widget file-input, requiring summarization for larger sets.
technicalImpact Medium · source · secondary source
Aug 2026
↓ Removes existing audio filtering capabilities from ElevenAgents configuration, potentially breaking workflows relying on these filters.
productImpact High · source · secondary source
Jul 2026
↓ Breaking change for API clients requiring schema updates to include conversation_id in WebRTC token responses.
technicalImpact Medium · source
Jul 2026
↓ Shorter default fallback window for the backup LLM cascade in ConvAI agents; remains user-configurable within the 2–15s range.
technicalImpact Low · source · secondary source
Jul 2026
→ Optional admin-set control: workspace admins can cap monthly credit consumption for new invitees. User-side budgeting capability, not a vendor-imposed limit.
technicalImpact Low · source · secondary source
Jul 2026
↓ Removes the ability to manually optimize streaming latency for TTS agents.
productImpact Low · source · secondary source
↔ shipped alongside New optional music model; music_v1 flow remains available. (Jun 2026) — more for more, not a pure cut.
Jul 2026
↑ Increased capacity for SIP REFER User-to-User Information data.
technicalImpact Low · source · secondary source
Jun 2026
→ New optional music model; music_v1 flow remains available.
productImpact Low · source · secondary source
Jun 2026
→ Dev-facing enum change: async option replaced by fixed time windows; existing configs need a one-time migration.
technicalImpact Low · source · secondary source
Jun 2026
↑ Batch agent-test capacity raised 1000 → 5000 attached tests.
technicalImpact Low · source · secondary source
Jun 2026
↓ Users must migrate to v2 models before July 9, 2026, or lose access to these specific models.
productImpact High · source · secondary source
↔ shipped alongside New optional music model; music_v1 flow remains available. (Jun 2026) — more for more, not a pure cut.
Jun 2026
↓ Users must migrate to scribe_v2 or scribe_v2_realtime before July 9, 2026.
productImpact High · source · secondary source
↔ shipped alongside New optional music model; music_v1 flow remains available. (Jun 2026) — more for more, not a pure cut.
Jun 2026
→ Newly added privacy feature is restricted to enterprise customers for text-to-dialogue endpoints.
policyImpact Low · source · secondary source
Feb 2026
→ v3 promoted to GA
productImpact Low · source
Sep 2025
↓ Two legacy TTS models removed; pinned integrations break
productImpact Medium · source
Jun 2025
↑ v3 launched with a temporary 80% discount (expired end of June)
businessImpact Low · source