OpenRouter

Evidence (18)

• 6The New Stack·10/10/2026product_launch
  Grok Bot will use the best model for each task

  Elon Musk said Grok Bot will route tasks to the best available backend model rather than stay loyal to a single model, including Claude Opus 5.5, MidJourney, Suno, and others. The article argues this reflects a broader shift in AI adoption, where organizations increasingly use model triage to control costs and improve results.

• 7The New Stack·10/9/2026model_release
  Microsoft launches Decision-1 for fast decision scoring

  Microsoft introduced Microsoft-Decision-1, a decision-scoring model designed for structured tasks such as routing, classification, prioritization, verification, and workflow control. The company says it offers strong latency and accuracy, and is available through Microsoft Foundry and OpenRouter.

• 7The New Stack·9/19/2026product_launch
  Zed, Anthropic, and OpenRouter push the AI harness layer

  This week’s coverage centered on tools and interfaces that make AI models usable in real workflows, rather than on new model releases. Zed expanded agentic code collaboration with Delta, Anthropic unified Claude Chat and Cowork, and OpenRouter launched in-region routing to give enterprises more control over data processing.

• 7The New Stack·9/18/2026research
  Open-weight models reach majority of tokens on Vercel AI Gateway

  Vercel reported that open-weight models handled 56% of tokens routed through its AI Gateway in August, the first time they have crossed a monthly majority. However, Anthropic still accounted for 64% of spend, showing that open-weight usage is growing faster than its share of revenue.

• 7The New Stack·9/14/2026product_launch
  OpenRouter adds US-only routing for business AI traffic

  OpenRouter has launched general availability of US in-region routing for business and enterprise users, ensuring requests sent through its US endpoint are decrypted, processed, and served entirely inside the United States or rejected. The move responds to growing use of open-weight models, especially Chinese models, in US token traffic and gives companies more control over data residency.

• 3Hacker News·9/14/2026research
  Updated LLM SVG benchmark re-run across six models

  A Hacker News post describes a follow-up to Simon Willison’s pelican-riding-a-bicycle benchmark, using newer LLMs to generate SVGs from similar prompts such as an octopus operating a pipe organ. The author tested ten prompts across six models on OpenRouter, found the exercise cost about $20, and shared the results on a linked site.

• 5Simon Willison·9/11/2026framework_update
  OpenRouter routing can change model behavior

  OpenRouter advertises automatic fallback routing and cost optimization across model providers, but different providers may run the same model with different serving software, settings, and capabilities. The article notes this can lead to inconsistent outputs, missing vision support, or different handling of reasoning options, and that users can restrict routing with the provider.only option.

• 6Hacker News·9/11/2026research
  Benchmark finds RTK token savings do not reliably reduce AI coding cost

  A benchmark analysis tested RTK, a terminal-output compression tool for AI coding agents, and found that large reported token reductions did not consistently translate into lower spend. Results varied by model and task: one setup saw a small cost improvement, while another became more expensive overall. The study concluded that less terminal output can help, do nothing, or even backfire depending on task structure and agent behavior.

• 4Hacker News·9/8/2026research
  Benchmarking model and harness combos on a Three.js task

  A Hacker News post reports tests of 10 model and harness combinations on the same prompt to build a single-page Three.js sci-fi hangar scene. The write-up compares runtime, token usage, tool calls, browser checks, and estimated costs across models such as GLM 5.3 Flash Max, GPT-5.6 variants, and Qwen 3.8 27B under different harnesses.

• 6Hacker News·9/4/2026product_launch
  GPT-6 Astra listed on OpenRouter

  OpenRouter now lists GPT-6 Astra, making the model available through its model marketplace and API aggregation layer. The Hacker News thread focuses on the model's presence on the platform and what that may imply for access and routing across providers.

• 7Hacker News·9/2/2026research
  Perplexity cites mass-generated software list pages from obscure domains

  A measurement of Perplexity/sonar and Perplexity/sonar-pro found that most citations for 380 software-category queries pointed to low-ranked or unranked domains, with a large share coming from newer sites. The study also identified three related domains that have published over 215,000 machine-generated “best <category>” pages and are being used as grounding sources by the models.

• 5The New Stack·9/1/2026model_release
  GLM-5.3-Flash Trades Time for Lower Cost

  The New Stack compared Z.AI’s GLM-5.3-Flash and GLM-5.3 on coding, reasoning, and information-extraction tasks. Both models produced correct results, but Flash used far more tokens and time while still costing less per request due to much lower pricing.

• 8Simon Willison·8/29/2026model_release
  Tencent Open-Sources Hy4 Preview LLM

  Tencent has released Hy4 preview, an open-source large language model with 770B total parameters, 49B active parameters, and a context window of over 1M tokens. The company says it is optimized for productivity tasks such as coding, office work, scientific research, and game development, and is available through Tencent products and APIs.

• 6Hacker News·8/27/2026open_source
  Show HN: Open-source model gateway that routes traffic to better models

  The project presents an open-source model gateway for managing self-hosted, frontier, and open-source models through one interface. It emphasizes low latency, support for many providers and models, and traffic-based optimization that uses traces and simulations to choose the best model for each request. The authors also say the system can be self-hosted or used as a hosted service with no markup.

• 8The New Stack·8/27/2026research
  Benchmarking Shows AI Coding Harnesses Can Drive 70x Token Differences

  Recent benchmarks compared multiple AI coding harnesses running the same underlying models and found large differences in token usage and cost. The results suggest that harness design, including prompt overhead and turn count, can matter as much as the model itself when pricing coding agents.

• 7The New Stack·8/27/2026product_launch
  Replit makes Auto mode the default for model routing

  Replit is rolling out its intelligent model routing system as the default for all accounts, automatically selecting models based on task needs, speed, and cost. Core and Pro users can still override Auto mode and choose models manually, while enterprise admins can limit routing to approved models.

• 8The New Stack·8/26/2026model_release
  Z.ai unveils GLM-5.3-Flash, a low-cost hybrid model on Chinese chips

  Z.ai revealed that the stealth model Ox-alpha is GLM-5.3-Flash, a 320B-parameter hybrid model with 18B active parameters designed for very low-cost inference. The model is open-sourced on Hugging Face under the MIT license and is already available on platforms like OpenRouter, with Z.ai emphasizing that it was served efficiently on Chinese AI chips using its own inference stack based on SGLang.

• 7The New Stack·8/24/2026model_release
  Ox Alpha Raises Privacy and Attribution Questions

  Ox Alpha has appeared as an anonymous coding model on OpenRouter and OpenCode, with strong speculation about who built it but no confirmed attribution yet. The bigger issue is that different access routes appear to carry different data-retention and training terms, creating uncertainty for developers sending private code through the model.