Inworld Realtime TTS-2 vs MiniMax Speech (T2A)

Inworld Realtime TTS-2 68

Read as a model. Reasoning and an updated inworld-tts-2 on 28 October 2026, and @inworld/cli 1.5.0 on 29 October (30). Dated TTS changes on 6 August, 9 August, 11 September and 28 October 2026 (10). inworld-tts-1 and -1-max were discontinued on 15 June 2026 with requests routed to newer models, and the 1.5 models are deprecated with no end date, but no notice period was found (5 of 10). Public release notes, a Discord server and support@inworld.ai, with no evidence read of how they answer (8 of 15). @inworld/tts 1.1.1 dates from 6 May 2026, before TTS-2 Flash and Reasoning, and the PyPI page answered with a bot check (10 of 15). The CLI is current, but the SDK and CLI repositories weren't read (5 of 10). Total 68.

MiniMax Speech (T2A) 38

Read as a model, as minimax-voice-cloning is. The newest dated change touching speech is the MCP servers' August update, minimax-mcp-js 0.0.18 on 21 August 2026, 77 days before 6 November, and Speech 2.8 dates from 23 January 2026 (20 of 30). No dated API entries in the last 90 days (0). Support by email and Discord, not tested (6 of 15). No official SDK for speech; the MCP servers expose text_to_audio and list_voices (8 of 15). The Python MCP server has tests and moved to version 2 of the MCP SDK in August, with no CI workflow (4 of 10). Total 38.

Inworld Realtime TTS-2 61

Model reading of the checklist, with training and retention in place of the least-privilege and injection lines. Workspace API keys, deleted to revoke, with separate keys per environment advised, single-use client tokens valid 15 minutes, and a Realtime-only key type. No key can be limited to TTS, because write toggles cover only the Voice and Router APIs (25 of 30). The Terms (11 June 2025) say Inworld won't train generally available models on customers' non-public Materials, except feedback and content flagged for trust and safety. The Privacy Notice (18 January 2026) lists training AI models among its uses and covers voice data, while excluding data processed for enterprise customers (12 of 20). Zero data retention is Enterprise only or a Growth add-on. No retention period is published for other plans, async results are kept about 7 days, and whether the Reasoning model's step is covered isn't stated (6 of 15). Usage is attributed per key, and a Performance page (Builder and above) shows errors, latency and throttling. The Logs page sits behind the Portal login and wasn't read (8 of 15). SOC 2 Type II, HIPAA and GDPR claimed on the security page, with the trust centre unread. A vulnerability report form exists. No security.txt (404) and no bug bounty found (10 of 20). Total 61.

MiniMax Speech (T2A) 36

Model reading of the checklist, with training and retention in place of the least-privilege and injection lines. Plain API keys with no scopes found, and leaked keys may be disabled (20 of 30). The terms (effective 30 March 2026), as read for minimax-voice-cloning on 4 November, let MiniMax use input and generated content to operate, maintain, develop and improve its services, with no opt-out found, and the privacy policy rules out only training to profile or target consumers (5 of 20). The speech docs call every interface stateless, and no retention period for submitted text or generated audio was found (5 of 15). Responses carry usage_characters and a trace ID, with no audit log (6 of 15). No security.txt, disclosure policy, bug bounty or certification found (0 of 20). Total 36.

Inworld Realtime TTS-2 60

No OpenAPI or other machine-readable file was found. llms.txt and the API reference don't link one and the usual paths redirect to pages (0). llms.txt, llms-full.txt, a Markdown twin of every page and a keyless docs MCP server (10). The models page says which model suits what, the Reasoning guide says where it runs and where it doesn't yet (WebSocket, Realtime, batch), and the reference marks fields each model ignores (17 of 20). Typed fields with enums for encoding, deliveryMode, reasoning and timestampType, a sample-rate range, per-endpoint text limits and a 600-character cap on reasoningContext (14 of 15). Examples in curl, Python, JavaScript and the CLI, with one 4XX example and the concurrency error, and no TTS error list (9 of 15). Versioned /v1 paths and dated TTS release notes, but the notes have no entry for the 28 October Reasoning launch and model update, and models have no dated snapshots (10 of 15). Total 60.

MiniMax Speech (T2A) 81

OpenAPI 3.1 for the sync and async routes and AsyncAPI files for the WebSocket and bidirectional sockets (25). llms.txt and a Markdown copy of every page (10). The text field says when to stream and how to mark pauses, and guides separate sync, WebSocket and async use (15 of 20). Enums for model and audio settings, with model and text required (13 of 15). Examples throughout and the shared table of error codes, but the OpenAPI file documents only a 200 response (10 of 15). Versioned /v1 paths. The API release notes stop at 28 October 2025, and the last T2A entry is 23 September 2025 (8 of 15). Total 81.

Inworld Realtime TTS-2 40

No x402, MPP or L402 (0). Per-character prices for both models and every plan are on the public pricing page, $25 per million characters for TTS-2 on the free plan (20). The On-Demand plan is free with up to 70 minutes of TTS, and the pricing table lists no payment method for it (20). inworld login opens a browser sign-in and the CLI mints keys only after it, so a person signs up (0). Total 40.

MiniMax Speech (T2A) 20

No x402, MPP or L402 (0). Per-character prices for every speech model published without a login (20). No free allowance for synthesis is published (0). A person signs up in a browser and funds a balance (0). Total 20.

Inworld Realtime TTS-2 62

Hosted reading of the checklist, on the TTS API. status.inworld.ai is a public status page running the issue-status app, which draws components and incidents from GitHub issues (20). The page is still a script shell on 4 November, and its history wasn't read, because GitHub's pages answer our proxy with 403, so no readable history (5). Concurrent generations, WebSocket connections and cloning and design requests a minute are published per plan, from 5 concurrent generations on the free plan to 500 on Growth (15). A 429 is documented with exponential backoff and jitter code, and the concurrency error (code 8) says which messages may have been dropped, but no Retry-After header is documented (12 of 15). An SLA is named on the Enterprise plan only, with no terms or uptime figure published (0). TTS-2 reached general availability on 31 August 2026 and Reasoning carries no Preview mark. Async and batch jobs are Preview (10). Total 62.

MiniMax Speech (T2A) 75

Atlassian Statuspage at status.minimax.io with a Text-to-Speech component (20). Its history, re-read on 6 November, shows three incidents in the 90 days from 8 August, all titled elevated errors and labelled major, 22 minutes on 28 September, 3 minutes on 16 October and 2 minutes on 19 October, and none in November. Elevated errors with no fuller scope are read as partial, as on minimax-api and minimax-voice-cloning, so minor only (20). T2A is limited to 60 requests a minute, with higher limits on audio subscriptions (15). Error codes 1002 and 2045 say to retry later and avoid sudden increases, with no Retry-After or backoff figures (10 of 15). No SLA (0). No beta or preview label on T2A (10). 20 + 20 + 15 + 10 + 0 + 10 = 75.

Inworld Realtime TTS-2 68

API reading of the checklist. Eight audio encodings from 8 to 48 kHz, timestamps only when asked, and synchronous, NDJSON streaming, WebSocket and async delivery. Reasoning adds one short string, reasoningInstruction, to the response (22 of 25). Text is capped at 2,000 or 4,000 UTF-16 units per request, and voices can be filtered by tags, but voice-list paging wasn't confirmed today (12 of 20). Errors carry a numeric gRPC-style code, a message and details, with one worked example and the code 8 behaviour explained, and no list (12 of 20). Synthesis is stateless and retry guidance is published, but the billing page says text is counted once synthesis starts, with no idempotency key (10 of 20). Three required fields, official Node.js and Python SDKs, the CLI, and an OpenAI-compatible endpoint that works with OpenAI's SDKs. The docs warn that SDK defaults may name older models (12 of 15). Total 68.

MiniMax Speech (T2A) 56

API reading of the checklist. Audio comes back hex-encoded inside JSON by default, with subtitles opt-in and format, bitrate and sample rate settable (15 of 25). Voices are listed by type through POST /v1/get_voice, which has no pagination and returns 300+ system voices for all (10 of 20). Errors come as numeric codes in base_resp with a short remedy for each, inside responses documented only as 200 (13 of 20). No idempotency key, and the MCP tools carry no read-only or destructive annotations (10 of 20). Two required fields. No official SDK for speech, with MCP servers in Python and JavaScript (8 of 15). Total 56.