OpenRouter

Two OpenRouter skills scaffold a coding agent you own. create-agent-tui builds a terminal harness with configurable tool display, input and banner styles; create-headless-agent builds a structured-IO agent for CLIs, API servers and queue workers. Both sit on the Agent SDK loop.

OpenRouter ran Opus 5.5, Sonnet 5.5 and Haiku 5.5 through identical coding tasks and a Claude Code bug fix. All passed, so cost separated them. It recommends Sonnet as default and Haiku for sub-agents.

OpenRouter's February roundup: benchmark scores from Artificial Analysis on every model page, an effective-pricing tab exposing tiered per-provider cost, benchmark scatter charts in rankings, and an openrouter/free router slug for free models.

A cost playbook for OpenRouter: append :floor to route to the cheapest provider for a model, set max_price so a request fails rather than overspends, use :nitro when throughput matters more, and check whether a cheap listing is actually a quantized endpoint.

Investigates whether Seedance 2.5/2.0 models have public weights or a license by checking ByteDance's own pages plus both ByteDance-Seed GitHub and Hugging Face orgs, finds none, and shows the OpenRouter video API call pattern to use instead.

OpenRouter adds streamed reasoning summaries for o3 and o4-mini through its chat completions API, an optional user field for per-customer tracking and moderation, a setting that pins a provider to your own key, crypto invoicing, 2FA and SDK usage accounting.

OpenRouter's live-tested comparison of embedding models by use case: text-embedding-3-small for English RAG, Voyage 4 for long inputs, Qwen3-Embedding for multilingual retrieval, Voyage Code 4 for code search, plus free and low-cost options.

Xiaomi's MiMo-V2.6 family is now live on OpenRouter: a 1T+ parameter flagship (Pro), an open-source MoE (Flash), and a ~10x faster UltraSpeed variant of Pro, all handling text, image, video and audio with 1M context.

Routed-token data showing an open-weight family roughly doubling its share in six months, driven by agentic rather than chat traffic. Includes the price gap against frontier models, adoption split across hobbyist and enterprise keys, and how agentic traffic was classified.

Compares TypeSafe's Jev with LLM-as-a-judge for grading agent outputs, measuring agreement, calibration, bias, latency, and cost, and says which judge fits closed versus open rubrics.

A month of gateway changes in one place: audio endpoints, parallel multi-model Model Fusion, private fine-tuned endpoints, a router that picks the cheapest model clearing a coding score, enforced IP allowlists, and per-provider zero data retention toggles.

OpenRouter opened a free stealth model, Ox Alpha, pitched at efficient coding and sustained agentic work. It offers a 1M token context and text, image and video input, and the provider says it will not train on prompts or completions. Developer identity and benchmarks are not disclosed.

Descript's Underlord video agent went from three hardcoded model integrations to 13 production models across four labs by routing inference through OpenRouter, cutting evaluation time from a week to under two hours.

OpenRouter walks through a small Python code reviewer built on its Responses API. It uploads changed files, runs them in a Linux sandbox to confirm each finding, returns patches, and reports cost and runtime across four models.

OpenRouter published Model Router Benchmarks comparing seven routers, including NVIDIA Switchyard and Unbiased Pareto, on quality, speed and cost across six benchmarks, plus model selection and reliability views.

OpenRouter lists OpenAI's Decisions API as GPT-6 Luna Decisions. It accepts text, JSON or images and returns typed answers with probabilities to pick a model, tool or action. Priced at $0.10/M input with free output and a 1M context window. Based on OpenRouter's post only.

OpenRouter added a /tools API endpoint that returns Server Tools Marketplace data, including best-in-market prices, so agents can consume it programmatically. The announcement is brief and the linked docs were not read, so schema details are unknown.

OpenRouter's US in-region routing is generally available alongside EU routing. Requests to the regional endpoints are decrypted and processed inside the chosen region at every step, from TLS termination through to the model provider. Linked docs were unreachable, so enforcement details are unconfirm

OpenRouter ran an identical prompt through 20 image generation models and read the real billed cost off each response, finding a 22x price spread and naming picks for cheapest, best editing, and text-in-image output.

OpenRouter walks through building a marketplace moderation system with the Jev API: structuring state, writing questions that survive literal reading, calling the TypeScript API, and calibrating publish/hold/reject thresholds from labeled data.

Five human-in-the-loop patterns for agents facing EU AI Act Article 14, Colorado SB26-189 and the NIST AI RMF: classify tools by risk tier, log every oversight event, escalate on timeout, back approval state with durable storage, and wire the gate into the Agent SDK loop.

OpenRouter's December roundup leads with Claude Code support for usage tracking and cost control, plus automatic JSON repair for structured outputs, browser notifications on long completions, and rankings filtered by context length.

PDF input works across all OpenRouter models, falling through to native handling for providers that support it and OCR for the rest. Two engines set the cost floor: mistral-ocr at $2 per 1,000 pages, or free text-only extraction with no image support.

The largest multi-model gateway is joining Stripe, with no announced change to the API, product or roadmap. Relevant mainly as dependency risk: routing neutrality across 400+ models is now a commitment made by a subsidiary rather than an independent company.