Worked check: Argon’s list rate is $2/$10 per million tokens through November 30, then $4/$20 — input and output are separate, and a loop pays both again on every step.

Inference is the bill and the machine. Argon's introductory rate is $2 per million input tokens and $10 per million output tokens through November 30, then $4 and $20 on Google's page. Muse Wonder is $4 and $20, with cached input at $0.20. Those are list prices for a call, not a verdict on a long loop, which pays both numbers again on every step.

Some models are not an API call. EmbeddingGemma 2 is small enough that search can stay on the device. Muse Spark Open, as Meta describes it, wants two M5 Ultra Mac Studios or a host. DeepSeek's DeepInfer repository is Apache 2.0 code for V4.1 Flash, and DeepSeek says the KV cache and the load balancer are not in that release. DeepSeek says one of its clusters serves 155 trillion tokens a day. The choice is which bytes move, and who pays for them.

In this briefing

  1. CUDA 13.4 can refuse a GPU allocation that crosses a memory limitOctober 27, 2026
  2. Liquid AI ships LFM3, a 1.7 billion parameter model that beats its larger predecessorOctober 27, 2026
  3. NVIDIA says the 4 GB Nemotron Bolt model is available, and uses fewer tokensOctober 27, 2026
  4. GPT-6.1 Terra is $1 and $5 per million tokens, and OpenAI's coding score more than doublesOctober 26, 2026
  5. Mistral prices Voxtral 3, including a 4B transcription model at $0.0035 a minuteOctober 20, 2026
  6. DeepSeek open-sources DeepInfer, and says one cluster serves 155 trillion tokens a dayOctober 20, 2026
  7. Mistral releases Timestral, a 209M model for forecastsOctober 15, 2026
  8. NVIDIA releases Nemotron 3.5 Super VL, a 120B open modelOctober 15, 2026
  9. Gemini 4 Argon opens to developers at $2 per million input tokens until November 30October 13, 2026
  10. Liquid AI releases a 24B agent model with 2B activeOctober 13, 2026
  11. EmbeddingGemma 2 searches text, images, audio, and video on the deviceOctober 6, 2026