OpenRouter

OpenRouter outlines a practical method for benchmarking LLMs
OpenRouter has published a guide to benchmarking large language models with application-specific tasks, code-based grading and repeatable runs. Its example study found…

OpenRouter launches beta localization in eight languages
OpenRouter has launched beta localization in seven languages alongside English, covering Chinese, Spanish, Brazilian Portuguese, German, French, Japanese and Korean. The…

OpenRouter calculates when self-hosting an LLM beats API costs
Self-hosting an open-weight LLM costs less than using an API only when traffic keeps the GPU busy enough, according to an OpenRouter analysis. Its calculations show that…

SambaNova brings Z.ai’s GLM-5.3 to SambaCloud at over 200 tokens/s
SambaNova says Z.ai’s GLM-5.3 is now available on SambaCloud at over 200 tokens per second, giving developers access to an open-weight model designed for long-horizon…

OpenRouter finds no clear quality gap between fp8, fp4 and fp16
OpenRouter found no detected accuracy difference between fp8, fp4 and fp16 in a test of Qwen3.8 27B, although the company cautions that the result covers one model and…

OpenRouter measures the token cost of MCP, CLI tools and skills
OpenRouter’s comparison of MCP, command-line tools and skills found that the way an AI agent receives a capability can dramatically change its context usage and cost. In…

Claude Opus 5.5 Leads OpenRouter’s Scanned Document OCR Test
Claude Opus 5.5 was the most accurate model in OpenRouter’s test of OCR for scanned documents, correctly identifying 262 of 264 key values. Mistral OCR delivered the…

OpenRouter launches Extractions API for typed data from documents
OpenRouter has launched the Extractions API in alpha, giving developers a way to pull named, typed fields from text, images and PDFs. The service returns structured…

OpenRouter launches Auto and Auto Code model routers
OpenRouter has launched a redesigned Auto Router and a new Auto Code router for coding work, using live request signals, pricing and provider data to choose models. The…

Sentry Retires 500 AI Keys After Moving Internal Use to OpenRouter
Sentry moved its internal AI usage to OpenRouter, retiring roughly 500 direct Anthropic and OpenAI keys and giving teams broader access to models under defined budgets.…

Developers are using more models and moving to their own hardware
Development teams are distributing work across an increasing number of model families and running more tasks on their own hardware. According to Kilo Code, costs…

Together AI outlines self-tuning approach to open-model inference
Together AI has outlined Polymorphic Inference, a self-tuning approach designed to adapt open-model serving to changing architectures, workloads and hardware. The…

Open-weight models hit 72% of top-50 token use on OpenRouter
Open-weight models handled 72% of the tokens among OpenRouter's top 50 models in late September 2026, up from 27% in October 2025, according to a new analysis from the…

TBC Launches mischief 1, an AI Video Model Built With Insights From Living Neurons
The Biological Computing Co. (TBC) has launched mischief 1, an AI video model it says generates video 5x faster than its base model. According to the company, it is the…

Claude Code Session Costs: Opus 5.5 Median $0.31, Sonnet 5.5 $0.16
Claude Code is free to install, but the model it calls is not. According to OpenRouter, a short session of 2 to 9 turns on an API key costs a median $0.31 with Claude…

OpenRouter rebuilds provider routing to reward speed and reliability, not only price
OpenRouter has replaced its price-focused default provider routing with a new algorithm that also weighs speed, throughput and reliability. According to the company…

OpenRouter to screen all inference requests for CSAM, including ZDR traffic
OpenRouter will screen all inference requests passing through its service for child sexual abuse material (CSAM), the company said in a blog post. Screening happens in…

MiniMax M3.1 Flash goes free in Kilo, pitched as a cheaper, faster M3
MiniMax M3.1 Flash, the company's newest model, is now available in the Kilo coding platform at no cost for a limited time. According to MiniMax, it delivers…

Krea 3 edits images in under 8 seconds at native 2K for $0.06
Krea has released Krea 3, an image editing model that, according to the company, returns most edits in under 8 seconds at native 2K resolution. It costs $0.06 per image…

Open-Weight Models Move Into Production, Kilo Panel Hears
Open-weight models are moving from experiments into production at banks, fintech and biotech firms, according to speakers at a panel hosted by Kilo (by Anaconda)…

Inworld launches inference service built for consumer AI apps
Inworld AI has opened Inworld Inference to every developer, describing it as the first inference service designed for consumer AI. According to the company, it delivers…

OpenRouter launches Private Endpoints for fine-tuned models and reserved capacity
OpenRouter has launched Private Endpoints, an add-on to its Enterprise plan that lets organizations route their own fine-tuned models, dedicated deployments and reserved…

n8n adds Gemini 4 Argon support for MCP, Assistant and Agents
Workflow automation platform n8n says it is one of the first companies to integrate Google's Gemini 4 Argon. According to n8n, users can apply the model in three ways…
ModelsMicrosoft launches Microsoft-Decision-1, a model built for fast decisions
Microsoft has introduced Microsoft-Decision-1, a model designed for fast decision-making on structured tasks. According to Microsoft CEO Satya Nadella, it outperforms…