Every change we detect, newest first. Times are Eastern (ET). Archive
What to do: Developers considering the model should weigh its above-average intelligence against slower output speed and high verbosity when estimating latency and token costs.
What to do: Call it as kimi-k3.1-pro-highspeed.
What to do: Use exact model IDs like k3.1-pro or k3.1-pro-highspeed, keep thinking enabled, and start new sessions when switching models or effort levels.
What to do: Call it as nemotron-3.5-bolt-e4b-4gb.
What to do: Call it as nemotron-3.5-bolt-e4b.
What to do: If you use Google Cloud, consider accessing the Claude API through your Google Cloud project and configure IAM access and Marketplace billing.
What to do: Review the Decisions API to see if it fits low-latency, low-cost decision workloads in your app.
What to do: Review the changes since public beta and update your Decisions API integrations to use the new features and 400-question limit.
What to do: Try streaming by creating a new gateway with the opt-in Preview flag pointed at an existing SSE backend, and implement client reconnection logic for streams that may end before the one-hour limit.
What to do: Try installing the CLI via Homebrew or Go, and if you use image variations, switch to images edit with a GPT Image model.
What to do: If you want self-hosted durable agents with your own models, try Omnara via its GitHub repo or hosted version.
What to do: Watch for availability in Gemini's Model Garden if you want to access OpenAI models through Google Cloud Marketplace.
What to do: Call it as nemotron-3.5-rerank-8b.
What to do: Call it as nemotron-3.5-embed-8b.
What to do: Call it as nemotron-3.5-embed-1b.
What to do: If you're on a Claude Enterprise plan, you can start using inference hooks in production.
What to do: Call it as voxtral-mini-tts-3-0.
What to do: Call it as voxtral-mini-transcribe-realtime-3-0.
What to do: Call it as voxtral-small-transcribe-3-0.
What to do: Call it as voxtral-mini-transcribe-3-0.
What to do: If you want to self-host DeepSeek models, try DeepInfer from source (Python 3.12, CUDA 13 or CANN 9.2) and plan to supply your own KV cache and load balancing.
What to do: If you're on Claude Enterprise and manage Slack Claude Tag settings, request a scoped Admin API key and try the beta endpoints with the experimental-claude-tag-2026-10-21 header.
What to do: Developers building agents who want full-trace error analysis without sampling can try flow-1 via Laminar or self-host the open-source platform.
What to do: If interested, request early access via the Maxwell site, or try the open-source optimised LTX-2.5 pipeline on your own workloads.
What to do: Upgrade to DeepSeek Harness 0.2.1-rc.1 or later for the best experience with the new model.
What to do: Try the prompt-to-voice endpoints to create custom voices, noting that speech generation is billed under regular GPT-Live session pricing.
What to do: Explore wiring Decisions into your Live sessions via native delegation or transcript delta triggers.