Nvidia released Nemotron 3 Diarization on September 23, an open-weight speaker-diarization model with roughly 100 million parameters that identifies up to eight speakers simultaneously at latency as l…
Seven inference providers document the single event of serving-capacity exhaustion under five different HTTP status codes — 429, 498, 500, 503 and 529 — according to a provider-by-provider review chec…
DeepSeek released DeepSeek V4.1 Flash on September 10, 2026, a sparse mixture-of-experts model the company reports exceeds its V4 Pro flagship on performance, speed, and task completion time. The mode…
Monitoring of the mistralai/mistral-nemo endpoint on deepinfra/fp8 was retired on 2026-09-06 after the B3IT border-input detector could not find enough border inputs, and the endpoint returns no logpr…
A new census of open-weight model hosting reveals that the same model name can arrive at different quantizations, context lengths, and output caps depending on the provider, with some hosts serving GL…
A federal judge ruled the Trump administration's blacklisting of Anthropic as a supply-chain risk was unconstitutional retaliation, citing First and Fifth Amendment violations, after Defense Secretary…
DeepInfra offers the cheapest API pricing for Qwen3 30B A3B at $0.12 per million input tokens, with Alibaba at $0.13 and NextBit at $0.14, according to live pricing data refreshed about 19 hours ago v…
Google's Gemma 4 26B A4B, a 25.2B-parameter multimodal mixture-of-experts model with 3.8B active parameters and 128 experts, is now available with DekaLLM offering the cheapest API pricing at $0.06 pe…
Cerebras' gpt-oss-120B (low) leads a new inference speed benchmark at 1,538.0 tokens per second, outpacing Groq's gpt-oss-20B (high) at 949.5 t/s and Cerebras' GPT-5.6 Sol (ultrafast preview) at up to…
A one-time snapshot of DeepSeek V4 Flash 0731 latency across nine inference providers found Baseten fastest with 3,980 decode tok/s and 7.72 s total p99, while Azure ran an older checkpoint and Scalew…
Ant Group's inclusionAI released Ling 3.0 Flash, an open-weights AI model that scores 38 points on the Artificial Analysis Intelligence Index, making it the smartest open model under 124 billion total…
A new Hugging Face dataset from hardik90/hf-inference-pricing-matrix reveals that the same AI model can cost up to 4.66x more depending on the inference provider, based on an analysis of 14 providers,…
DekaLLM offers the cheapest inference for Mistral Nemo 12B at $0.02 per million input tokens, delivering 3,444 points per dollar based on a general score of 62. The 12.2-billion-parameter model suppor…
DeepSeek V3 0324, a 671B-parameter Mixture-of-Experts model with 37B active per token and a 128k context window, is available on DeepInfra at $0.24 per million input tokens, the cheapest listed provid…
Morph offers the cheapest API pricing for Qwen3.6 27B at $0.29 per million input tokens, according to live provider data refreshed about one hour ago via OpenRouter. The 27-billion-parameter model, re…
A new measurement study of prompt cache keepalive economics across Anthropic, OpenAI, Gemini, and DeepSeek finds that the common practice of pinging every 30 seconds costs 8x more than necessary, and …
Meta is reportedly launching Meta Compute, a cloud service offering hosted Llama model APIs and raw GPU rental, directly competing with AWS, Azure, and Google Cloud. Bloomberg reported the plans on Ju…
ONNX v1.22.0, published June 30, introduces Opset 27 with LinearAttention and CausalConvWithState operators that enable modern hybrid LLMs to export cleanly to ONNX for the first time, alongside WebAs…
DeepInfra, a Palo Alto-based GPU inference platform, opened a 1.7MW data center in Toronto on July 8, housing more than 1,000 NVIDIA Blackwell B300 GPUs in its first deployment outside the United Stat…
DeepInfra opened a 1.7 MW Toronto data center on July 8, its ninth site and first outside the United States, hosting over 1,000 NVIDIA Blackwell B300 GPUs for low-latency AI inference. The expansion s…