Updates

Google released gemini-nano-banana-2.1 as a stable image model on October 6, 2026. gemini-3.1-flash-image is deprecated.

OpenAI API usage tiers are now Build ($5), Launch ($100) and Grow ($500 in credit purchases). Upgrades happen automatically.

Anthropic released Claude Haiku 5.5 (claude-haiku-5-5) on October 7, 2026. 1M context, 128k output. Code written for Haiku 4.5 can break.

New /v1/decisions endpoint in public beta. Works only with gpt-6-luna. Returns probabilities, choices or scores. $0.10 per 1M input tokens.

On October 7, 2026 Anthropic cut Claude Sonnet 5.5 prompt cache reads from $0.20 to $0.10 per million tokens. Other prices are unchanged.

Google deprecated gemini-3.7-flash and gemini-3.5-flash on October 8, 2026. deep-research-pro-preview-12-2025 shuts down October 23, 2026.

Deepgram's nova-3-medical model now accepts language=multi, covering ten languages in medical audio, for batch and streaming.

Firecrawl /scrape now reads PDFs, images, Word files, spreadsheets and presentations by URL. PDF parsing costs 1 credit per page.

Pinecone Local, the Docker emulator for local testing, is no longer available. Pinecone says to test in a separate project.

Deprecated on Sep 30, 2026. After Nov 30, requests to claude-sonnet-4-5-20250929 fail. Anthropic recommends moving to claude-sonnet-5-5.

Paid input goes from $0.75 to $1.50 per 1M tokens and output from $3.75 to $7.50. The current price is listed only through Dec 31, 2026.

New mistral-large-4 model, half price for its first two weeks. Open weights are promised soon. The discount ends around October 20.

The four framework adapters in @supabase/server are deprecated and will be removed on Dec 1, 2026. Supabase points to its framework bridges.

Subtitles (.vtt), Markdown, iOS .strings, YAML, .properties, .resx, .xlsm, .odt, .rtf and SCORM .zip can now be uploaded for translation.

New eleven_v4 model: more expressive speech and better voice cloning in 90+ languages. eleven_v4_turbo answers in about 100 ms.

Put a pause marker in your text and Flux TTS waits. Spell a word in IPA and it says it your way (Early Access).

The official Node SDK now accepts grok_imagine_1_5_lite for text-to-video and image-to-video. Upgrade the package and swap the model.

Since 2026-09-22 gemini-3.8-flash-tts and flash-lite-tts are GA, with 150+ voices, voice design and replication.

OpenAI shipped GPT-6 Sol and the low-cost GPT-6 Luna on 2026-09-22 for the Responses and Chat Completions APIs.

Google made gemini-3.8-flash GA on 2026-09-02: its most capable Flash model, with 1M-token input and function calling.

langchain v1.4.0 (Sep 1, 2026) adds langchain.mcp: load any MCP server's tools into create_agent.

Since Sep 2, 2026, Pinecone indexes can mix BM25 keyword, dense and sparse fields in one schema. Exact-match SKUs and error codes.

New nova-3-pharma model tuned for pharmaceutical vocabulary and drug names; English, batch and streaming. Swap one model param.