Back to Blog
blog••
0

AI model ecosystem briefing — 2026-11-21

Unattended daily digest of the LLM API landscape: 4 releases, 2 new models, 19 pricing updates and 1 community signal.

1. Today at a glance

  • Models tracked: 476
  • New models: 2
  • Pricing updates: 19
  • Community signals: 1

This briefing is generated by the unattended AINode content engine from public model catalogs, GitHub releases and developer community discussions.

2. New models

Model IDProviderContextInput /1MOutput /1M
thinkingmachines/inkling-2-mini thinkingmachines 1.0M $0.100 $0.400
radixark/daisyrose-27b radixark 262K $0.150 $1.88

3. Pricing updates

  • nvidia/nemotron-3.5-super-vl output $0.500 → $0.470 (−6%)
  • ~deepseek/deepseek-pro-latest input $0.119 → $0.100 (−16%)
  • deepseek/deepseek-v4.1-flash input $0.300 → $0.050 (−83%)
  • deepseek/deepseek-v4.1-flash output $1.20 → $0.199 (−83%)
  • deepseek/deepseek-v4-flash-vision-exp input $0.440 → $0.216 (−51%)
  • deepseek/deepseek-v4-flash-vision-exp output $1.32 → $0.647 (−51%)
  • google/gemma-4-26b-a4b-it input $0.090 → $0.068 (−25%)
  • google/gemma-4-26b-a4b-it output $0.300 → $0.225 (−25%)
  • minimax/minimax-m2.7 input $0.210 → $0.600 (+186%)
  • minimax/minimax-m2.7 output $0.840 → $1.80 (+114%)
  • minimax/minimax-m2.5 input $0.295 → $0.270 (−8%)
  • minimax/minimax-m2.5 output $1.20 → $0.950 (−21%)

4. Official & open-source releases

  • ollama/ollama v0.40.17:v0.40.17
    • ollama launch claude supports Claude Code's deferred tool loading, so MCP tools are loaded when needed instead of all at once
    • Ollama on Apple silicon now serves multiple requests at once.
    • ollama launch opencode sets thinking options from the model's metadata
    • ollama launch pi refreshes Pi's model list after sign-in and passes along prompt arguments
    View release ↗
  • vllm-project/vllm v0.34.0:v0.34.0
    • 10 new models (Mistral Large 4, MiniMax-M3.1-Flash, Nemotron-3.5-Transcribe, and more!)
    • Extensible KV cache is on by default: vLLM sizes the KV cache from memory actually free after warmup, so most models get a bigger cache and more throughput - in most cases you should now remove gpu_memory_utilization from your configs.
    • Model Runner V1 is gone. Model Runner V2 is now the only GPU runner.
    • Big speedups on Blackwell and MI355X: up to 1.95× decode (Step5 on GB200), 1.6× throughput (GLM-5.3-Flash), 1.24–1.40× faster long-prompt TTFT (GLM-5.3).
    View release ↗
  • deepseek-ai/DeepSeek-V3 v1.0.0:v1.0.0 View release ↗
  • meta-llama/llama-models v0.2.0:v0.2.0 View release ↗

5. Community pulse

6. Gateway tip

Keep one OpenAI-compatible endpoint and one key across providers. Route failover, meter tokens and bill by model without touching provider SDKs.