SilentAPI 0.9.10

October 27, 2026

Models and chat

Added Gemini 4 Argon, GPT 6.1 Terra, Claude Haiku 5.5, Kimi K3.1 Pro, Mistral Large 4, MiniMax M3.1 Flash, Muse Wonder and Muse Spark Open.
Updated DeepSeek checkpoints and kept checkpoint numbers in descriptions rather than model names.
Older GPT 6 Sol, Claude Haiku, GLM 5.2 and MiniMax M3 IDs resolve to their replacements.
Added web search to GPT 6 chat models, with the same search-call billing for JSON and streaming responses.
Fixed GLM 5.3 Flash routing reliability. Reasoning choices follow each model's supported controls, and chat now uses the model's published default.
Kimi K3.1 Pro is the default chat model. MiniMax text models appear under M3.

Added Nano Banana 2.1, Grok Imagine Video 1.5 Lite and Voxtral 3.0 speech and transcription models. Earlier Voxtral Mini TTS and Mini Transcribe models are no longer available.
Updated cache and long-context pricing, including Haiku's separate one-hour cache writes.
Fixed image-token billing when a response omits a total upstream cost. Grok image charges now follow the selected quality, resolution and input-image count.
Applied MiniMax video reference-image allowances before charging for extra references. Unsupported video and audio inputs return an error rather than being ignored.
Removed Maps grounding. Web search remains available on supported models.

Privacy and precision labels reflect each model's configured route. SilentAPI protected models remain distinct from End-to-end ZDR models.
Updated the privacy policy and terms to explain flagged-prompt retention exceptions and privacy considerations for sensitive tasks.
Updated Services, calculator, chat, API documentation and model lists. Platform service fees and native Cloud Vision service fees remain unchanged.