Additive changes only
FastInfra API changelog
The FastInfra API changelog records every change that affects integrators: new endpoints and models, rate-limit increases, compatibility notes and billing updates, each dated and listed before it ships. Changes are additive by policy: existing request and response shapes are not removed, field names do not change, and rate limits only ever go up. Read it before upgrading an SDK, and subscribe to the blog feed for longer write-ups of the larger releases.
Additive changes only. Anything that alters limits or behaviour for existing integrations is announced here before it ships. Limits only go up.
Text-to-speech: 2.5x Chatterbox throughput, strict FIFO, streaming, async jobs, Kokoro fixes
13 October 2026- Chatterbox Multilingual now runs on four GPU workers under CUDA MPS: 4.9x real time aggregate (was 1.9x on one serial worker). A 30-minute chapter renders in about 6 minutes instead of 15.
- Requests are served in strict arrival order (one per worker, the rest queued FIFO), so chunks submitted in sequence come back in sequence. Every response carries X-Queue-Position, X-Queue-Wait-Seconds and X-Speech-Worker; a full queue answers 503 with a real Retry-After instead of making everyone late.
- "stream": true on POST /v1/audio/speech for both models: chunked WAV or PCM, first audio in ~0.1-0.4 s (Kokoro) or ~1-2 s (Chatterbox). Upstream timeout raised from 120 s to 300 s for buffered calls.
- POST /v1/audio/speech/jobs for chapter-length work: 202 + job id immediately, GET .../jobs/{id} for status and queue position, GET .../jobs/{id}/audio for the file (kept 20 minutes). Same body, same price, billed once.
- Kokoro: fixed bursts of 503/400 (UnicodeDecodeError) and silently inconsistent output under concurrent load (its G2P library is not thread-safe and is now serialised), and fixed truncation of long paragraphs without newlines - text is now split on sentence punctuation including Devanagari danda. X-Speech-Truncated: 1 flags the rare segment that still hits the model's limit.
- Kokoro now speaks Spanish, French, Italian, Brazilian Portuguese and Mandarin Chinese as well as English and Hindi: pass a voice id with that language's prefix (ef_dora, ff_siwis, if_sara, pf_dora, zf_xiaobei, ...). Same $0.30 per 1M characters.
- /status.json lists Whisper, Kokoro and Chatterbox separately under components, with the live Chatterbox queue depth and worker count.
Chatterbox Multilingual TTS in 23 languages, and TTS is now 50% cheaper than anywhere else
12 October 2026- New text-to-speech model resembleai/chatterbox-multilingual (aliases chatterbox, tts-1-hd) on POST /v1/audio/speech: Resemble AI's open-source 0.5B neural voice with zero-shot voice cloning in 23 languages (Arabic, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish). Pass "language": "hi" etc.; same JSON-in, WAV-out contract as tts-1.
- Pricing: Chatterbox Multilingual $0.50 per 1M characters (DeepInfra lists it at $1.00). Kokoro-82M (tts-1) drops from $15 to $0.30 per 1M characters (DeepInfra $0.62, Together AI $4.00). Both run on FastInfra's own EU GPU pods; no third-party TTS API is called.
- Both models are listed in GET /v1/models and on the Models pages. Existing Kokoro calls are unchanged: same voices, same endpoint, lower bill.
- The live-avatar pilot (POST /v1/avatars/sessions) has been retired; the GPU now serves Whisper, Kokoro and Chatterbox.
Rate limits rebuilt around paying customers
27 September 2026- Paid accounts: 600 requests/min per key and 1,800/min per account (was 120 and an undocumented 180). Admin and internal accounts always get paid limits.
- The per-IP limit is gone. Several sites on one server never share a quota.
- Token budgets (1M tokens/min per key on chat) replace request counts as the real protection for GPU capacity; they never bind before the request limit at normal sizes.
- Every inference response now carries x-ratelimit-* headers (OpenAI/Groq names) with Together-style aliases, so SDKs pace themselves.
- 429 bodies carry a code naming the limit, the real Retry-After in seconds, your accepted and rejected counts for the last minute, and a plain statement when a client is retrying in a loop.
- Paying customers are admitted ahead of free-tier accounts when the server is near capacity. Published limits never shrink.
- A sustained storm of rejected requests now alarms us and emails the account owner within minutes.
Tool calling, embeddings, honest errors
27 September 2026- Tool calling round-trips correctly: tool_calls on assistant messages, tool_call_id on tool messages, finish_reason: tool_calls, and streamed delta.tool_calls with index.
- POST /v1/embeddings on self-hosted models at $0. text-embedding-3-small/-large and text-embedding-ada-002 are accepted as aliases; encoding_format: base64 and dimensions are supported.
- Every 4xx error carries a stable code and a docs_url. 401 and 402 responses include the URL to act on (api_keys_url, topup_url) and the model and minimum top-up involved.
- New Limits page in the dashboard: your tier, every limit, live headroom per key, 2xx vs 429 per hour, and a filterable request log with CSV export.
- Public status page at /status with per-model latency, 24-hour uptime, and gateway p50/p95.
Frequently asked questions
Does FastInfra make breaking API changes?
No. Changes are additive: new fields, endpoints and models are added, and limits only increase. Anything that would alter behaviour for an existing integration is announced here first.
How do I find out when a new model is added?
New models appear in the model catalog and GET /v1/models automatically; larger additions are noted here and on the blog.
Where do I report a compatibility problem?
Open a support ticket from the console. Compatibility issues with the OpenAI request format are treated as bugs.
Is there a feed for the changelog?
The blog RSS feed at /feed/blog.xml carries release write-ups; the changelog page itself is the canonical dated list.
Last reviewed October 15, 2026 by FastInfra. Prices and counts on this page come from the live catalog.