April Release Spotlight

April’s biggest releases: video generation, workspaces for multi-project isolation, and a TypeScript SDK that turns any model into an agent. Here’s what shipped.

Video Generation

Generate video from text or images through one unified API. We’re supporting Seedance 2.0, Veo 3.1, Wan 2.7, Sora 2 Pro, and more on day one, with normalized parameters, async job tracking, and capability discovery built in. Video sits alongside text, images, audio, embeddings, and rerankers under the same routing and billing layer.

Workspaces

Organize your OpenRouter projects into separate environments, each with its own API keys, routing defaults, guardrails, and observability. Built for multi-project developers, enterprise teams, and agents that need staging vs. production isolation.

Your existing setup lives in a Default workspace. If you don’t need multiple environments, nothing changes.

Agent SDK with callModel

@openrouter/agent gives you callModel: one function that handles multi-turn tool calling, streaming, stop conditions, and cost tracking across 400+ models. Define tools with Zod schemas, set stop conditions like stepCountIs(10) or maxCost(1.00), and let the SDK run the agentic loop.

Two skills ship on top of it: create-agent-tui to scaffold a coding agent with a terminal UI, and create-headless-agent to build a multi-model CLI tool with any coding agent.

Also shipped this month

Reranker models. A new modality for re-ordering search results and document chunks by relevance. Cohere Rerank and Fireworks rerankers are live with a new /rerank endpoint. API docs · Models

Model Fusion. Fuse the results of multiple models into a single response. Try it

Prompt history. View, replay, and remix previous prompts from OpenRouter. Audit output quality, iterate on prompts, and replay the same input across different models for comparison.

Benchmarks on model compare. Model comparison pages now show Design Arena ELO rankings with visualizations for 3D, website building, SVG, and more. Click “Compare” at the top of any model page. Compare models

Knowledge cutoff dates. LLM training cutoff dates are now available in the /models API, so you can programmatically check how current a model’s knowledge is.

Stripe Projects. Run stripe projects add openrouter/api to get an OpenRouter account, an API key, and Stripe billing, all from the command line. Your agents can do it too. Announcement

Typed tool context in the TypeScript SDK. Define a contextSchema on tools, pass context from callModel, mutate it with setContext(), and changes persist across turns, all Zod-validated.

Frontier model drops

Massive model launches this month:

• GPT-5.5 & GPT-5.5 Pro: OpenAI’s newest frontier models, state-of-the-art for long-running work across code, data, and tools
• DeepSeek V4 Pro & V4 Flash: Huge jump over V3.2, meeting or surpassing current state-of-the-art across benchmarks
• Claude Opus 4.7: Anthropic’s most capable Opus, built for long-running async agents
• GPT-5.4 Image 2: GPT-5.4 with state-of-the-art image generation
• Kimi K2.6: Moonshot AI’s long-horizon coding model built for sustained agentic work

Are we missing something you want to see? Let us know on Discord.