Delivered-To: tjphuhs@gmail.com Received: by 2002:a05:6000:298f:20b0:451:8158:aa77 with SMTP id gl15-n2csp1071106wrb; Thu, 7 May 2026 09:13:00 -0700 (PDT) X-Received: by 2002:ad4:5692:0:b0:8bd:6baa:6a0f with SMTP id 6a1803df08f44-8bd6baa72f4mr53939576d6.38.1778170380574; Thu, 07 May 2026 09:13:00 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1778170380; cv=none; d=google.com; s=arc-20240605; b=B/zM3HdXhH+FYhP5VTRYYvrbpVCZ5Do/mvBUtfT3/xF/WAoZJftIFqQtOlo5jZHNug g6Sq6Y6vklWQ4Xu0dDt1jsstIxepiSfdNjOuoxSvYZR263ZNe6ox+m6l3bJ8MvrTiu9p 848lEQ4LF2DoY5BD4alv0vGefumYE+uX4S2QM1x2BkvLTqMRsnFznDJ7UizISYOnAFpo GsOe3IYgtaM5+SicSRAutpLBM3ushy/gfT1mWeFu2yHVtBFaXRF6mX2biz+1JBkWXlGq gSX8vUTrS49X343yr9nhpBSOc91083EAXFUjM2auCybgJzQApCpt+z31GzrhmVdiZUCe FqUw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20240605; h=message-id:list-unsubscribe-post:list-unsubscribe:to:from:subject :mime-version:date:dkim-signature:dkim-signature; bh=j5rYECC4btbKrUdMJJ3SlW0RVPuvESEtwtAcXdC5YkY=; fh=9zP21OW67FFZD2VV42mGudo540Qp4aalMWxJB5kMYug=; b=F96F3rEtjBzTykApXKO2gSQVuOvFmPuLnA6SmwHWWTWZrGJZxh+0AZRQ4EenKyZDmn Bl3cq6UAbYM9aT9WoC5JEAb4BuVfF/TjKviP1vWEs3SGd9IiIs26MySbPFJcsTf5STtQ EQ9tJolOiryTd5IERNQIU9RQqu/OXGBZluWKAan056x13GrYGrwVhFRc/dLr27NeQ2L/ R/K+EbxHa4oK3GnGVX6oHQjIdhtJm3opztRvumqQ90LeEmwWbWy+RY3fTeE48PJ/ETf1 JE5eLC4fxkHVZ9vYDF3T3fykqBrxOHFpzSdogvC120R75Q7et59JQi8q8oDq+ZFChUq/ sWVQ==; dara=google.com ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@cio121019.openrouter.ai header.s=mailo header.b=YmsQUDKz; dkim=pass header.i=@mailgun.org header.s=mg header.b=ZFk0pyoY; spf=pass (google.com: domain of postmaster@cio121019.openrouter.ai designates 143.55.233.8 as permitted sender) smtp.mailfrom=postmaster@cio121019.openrouter.ai; dmarc=pass (p=QUARANTINE sp=QUARANTINE dis=NONE) header.from=openrouter.ai Return-Path: Received: from v58.v596efc78.use4.send.mailgun.net (v58.v596efc78.use4.send.mailgun.net. [143.55.233.8]) by mx.google.com with UTF8SMTPS id 6a1803df08f44-8b538459b8bsi236252966d6.100.2026.05.07.09.13.00 for (version=TLS1_2 cipher=ECDHE-ECDSA-AES128-GCM-SHA256 bits=128/128); Thu, 07 May 2026 09:13:00 -0700 (PDT) Received-SPF: pass (google.com: domain of postmaster@cio121019.openrouter.ai designates 143.55.233.8 as permitted sender) client-ip=143.55.233.8; Authentication-Results: mx.google.com; dkim=pass header.i=@cio121019.openrouter.ai header.s=mailo header.b=YmsQUDKz; dkim=pass header.i=@mailgun.org header.s=mg header.b=ZFk0pyoY; spf=pass (google.com: domain of postmaster@cio121019.openrouter.ai designates 143.55.233.8 as permitted sender) smtp.mailfrom=postmaster@cio121019.openrouter.ai; dmarc=pass (p=QUARANTINE sp=QUARANTINE dis=NONE) header.from=openrouter.ai DKIM-Signature: a=rsa-sha256; v=1; c=relaxed/relaxed; d=cio121019.openrouter.ai; q=dns/txt; s=mailo; t=1778170380; x=1778177580; h=Message-Id: List-Unsubscribe-Post: List-Unsubscribe: To: To: From: From: Subject: Subject: Content-Type: Mime-Version: Date: X-Feedback-Id: Sender; bh=j5rYECC4btbKrUdMJJ3SlW0RVPuvESEtwtAcXdC5YkY=; b=YmsQUDKz88/+4J0fwCEADKRbvV/f7WcmDPLuZKAj4D5wazLEnDwSA85eQDBQX3wSske0D3arQhW//rWhlFmdmON9wm8Gtfe3edCtYJk9T/CMUSfLpucsQNvltsf65dYgq173dAzekzZF0noOX0ZpQfhUSnHWnbqbotteGQkGBYI= DKIM-Signature: a=rsa-sha256; v=1; c=relaxed/relaxed; d=mailgun.org; q=dns/txt; s=mg; t=1778170380; x=1778177580; h=Message-Id: List-Unsubscribe-Post: List-Unsubscribe: To: To: From: From: Subject: Subject: Content-Type: Mime-Version: Date: X-Feedback-Id: Sender; bh=j5rYECC4btbKrUdMJJ3SlW0RVPuvESEtwtAcXdC5YkY=; b=ZFk0pyoY/QdQbrVBEz1LO+90ETsUJmwDE3zLQubQ18GumOVkcKx/Jpv2JWdSb9gJaROpmKAqFu5yJu37YufHv4xCywHGC/dX9qjZBhnF0DFaGLMNe88wXKNiJPIS0VhHGX7jcgAQc97+YBbM/Lb5xwGYefTTmbggsxpxfPr5H1g= X-Mailgun-Sid: WyIzYWIyNyIsInRqcGh1aHNAZ21haWwuY29tIiwiNDhjMjA1Il0= X-Feedback-Id: postmaster@cio121019.openrouter.ai::67ed41d425427d3a0d318d12:mailgun Received: by f8cd214c3fe7df4221bd02975fcb8ea5963d55a9039a5649a782d9a301fc4f42 with HTTP id 69fcba0c8073340fc22f151b; Thu, 07 May 2026 16:13:00 GMT X-Mailgun-Sending-Ip: 143.55.233.8 Date: Thu, 07 May 2026 16:13:00 +0000 Mime-Version: 1.0 Content-Type: multipart/alternative; boundary="8b5956134fd841ece819830beee882642d3faa0d3520dd712e092e51e95a" Subject: Video generation, workspaces, 35 new models. Here's what shipped. From: OpenRouter Team To: tjphuhs@gmail.com X-Mailgun-Dkim: true X-Mailgun-Native-Send: true X-Mailgun-Track-Clicks: false X-Mailgun-Track-Opens: false X-Report-Abuse-To: badactor@customer.io List-Unsubscribe: , List-Unsubscribe-Post: List-Unsubscribe=One-Click X-Mailer: Customer.io (dgSE_QoDAIvZ9gKK2fYCAZ4DMOtv_Kglb503qIvl1Q==; +https://whatis.customeriomail.com) X-Mailgun-Variables: {"email_id": "dgSE_QoDAIvZ9gKK2fYCAZ4DMOtv_Kglb503qIvl1Q=="} Message-Id: <20260507161300.dac59027bcc51b6c@cio121019.openrouter.ai> --8b5956134fd841ece819830beee882642d3faa0d3520dd712e092e51e95a Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" Now generate video from text or images through one unified API. Use a TypeS= cript SDK to turn any model into an agent. Organize your OpenRouter project= s into separate environments with Workspaces. =C2=A0 OpenRouter OpenRouter ( https://openrouter.ai?utm_campaign=3D050726_monthly-recap_s1&u= tm_content=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_so= urce=3Dcustomer.io ) ********************** What shipped in April. ********************** Now generate video from text or images through one unified API. Use a TypeS= cript SDK to turn any model into an agent. Organize your OpenRouter project= s into separate environments with Workspaces. And 35 new models landed. Her= e's the recap. Platform features 01 Video generation Video APIs are fragmented. Providers use different request shapes, paramete= r names, and billing units. We built one API around those differences. Submit a prompt to /api/v1/videos, get a job ID back, retrieve the video wh= en ready. Parameters are normalized across every model (resolution, duratio= n, aspect ratio, audio, frame images, reference images). A capability-disco= very endpoint tells you what each model supports before you call it. sora-2-pro veo-3.1 seedance-2.0 wan-2.7 Browse video models =E2=86=92 ( https://openrouter.ai/models?fmt=3Dcards&ou= tput_modalities=3Dvideo&utm_campaign=3D050726_monthly-recap_s1&utm_content= =3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source=3Dcus= tomer.io )|Try the demo =E2=86=92 ( https://multimedia-explorer.openrouter.= ai/?utm_campaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-= +V3+Final+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) 02 Workspaces Separate environments for separate projects, each with its own API keys, ro= uting defaults, guardrails, and observability dashboard. Useful for keeping= side projects, prototypes, and production cleanly isolated under one accou= nt. Create a workspace =E2=86=92 ( https://openrouter.ai/workspaces/new?utm_cam= paign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B= &utm_medium=3Demail_action&utm_source=3Dcustomer.io )|Read the docs =E2=86= =92 ( https://openrouter.ai/docs/guides/features/workspaces?utm_campaign=3D= 050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_med= ium=3Demail_action&utm_source=3Dcustomer.io ) 03 Agent SDK with callModel One function handles multi-turn tool calling, streaming, stop conditions, a= nd cost tracking across 300+ models. Stop conditions are composable: cap a = session at Cost(1.00), stop when a specific tool gets called, limit by turn= count, or combine them. The @openrouter/agent TypeScript package ships wit= h two scaffolds: create-agent-tui for a terminal harness and create-headles= s-agent for a server-side one. Typed tool context with Zod is supported. SDK docs =E2=86=92 ( https://openrouter.ai/sdk?utm_campaign=3D050726_monthl= y-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_a= ction&utm_source=3Dcustomer.io )|Build your own harness =E2=86=92 ( https:/= /openrouter.ai/announcements/create-agent-harness-with-agent-sdk?utm_campai= gn=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&ut= m_medium=3Demail_action&utm_source=3Dcustomer.io ) 04 Benchmark scores on every model page Every model page now shows benchmark scores from Artificial Analysis: codin= g, math, science, and long-context reasoning. Useful when deciding between = models at similar price points, or weighing open-weights against proprietar= y. The Rankings page ( https://openrouter.ai/rankings?utm_campaign=3D050726= _monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3D= email_action&utm_source=3Dcustomer.io ) picked up the same data with scatte= r charts and long-context trends from 100K to 1M tokens. See the rankings =E2=86=92 ( https://openrouter.ai/rankings?utm_campaign=3D= 050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_med= ium=3Demail_action&utm_source=3Dcustomer.io ) 05 I/O loggingBETA The Activity log can now capture the full request and response for every AP= I call. Open any generation to inspect the prompt, switch to Dev mode for r= aw JSON, or replay it in chatroom against a different model. Enable I/O logging =E2=86=92 ( https://openrouter.ai/settings/observability= ?utm_campaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3= +Final+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) Also shipped in April Reranker models with Cohere and Fireworks, Model Fusion for combining outpu= ts across multiple models, prompt history with replay and side-by-side comp= arison, Design Arena ELO rankings on the comparison pages, knowledge cutoff= dates exposed in the /models API, Stripe Projects for streamlined account = setup, and typed tool context with Zod validation in the TypeScript SDK. New models Kimi K2.6 moonshotai/kimi-k2.6 =C2=B7 Apr 20 1.82T tokens since launch. 1T-parameter MoE, 32B active per token. SWE-Benc= h Pro 58.6% beats GPT-5.4 and Claude Opus 4.6. Open weights on Hugging Face= =2E $0.74/M in$3.49/M out262K ctxopen weights View on /models =E2=86=92 ( https://openrouter.ai/moonshotai/kimi-k2.6?utm_= campaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Fina= l+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) Claude Opus 4.7 anthropic/claude-opus-4.7 =C2=B7 Apr 16 935B tokens since launch. Anthropic's frontier reasoning and coding model. = Beats Opus 4.6, GPT-5.4, and Gemini 3.1 Pro across reasoning and code bench= marks. $5/M in$25/M out1M ctx View on /models =E2=86=92 ( https://openrouter.ai/anthropic/claude-opus-4.7= ?utm_campaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3= +Final+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) GPT-5.5 + GPT-5.5 Pro openai/gpt-5.5 =C2=B7 Apr 24 OpenAI's latest. Pro variant pushes harder on the reasoning ceiling at the = cost of latency and price; the base model is the workhorse. $5/M in$30/M out1M ctxPro: $30/$180 per M View on /models =E2=86=92 ( https://openrouter.ai/openai/gpt-5.5?utm_campai= gn=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&ut= m_medium=3Demail_action&utm_source=3Dcustomer.io ) deepseek/deepseek-v4-flash ( https://openrouter.ai/deepseek/deepseek-v4-fla= sh?utm_campaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+= V3+Final+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) 704B tokens since launch =C2=B7 1M ctx =C2=B7 $0.14/$0.28 per M =C2=B7 Apr = 24 deepseek/deepseek-v4-pro ( https://openrouter.ai/deepseek/deepseek-v4-pro?u= tm_campaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+F= inal+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) 1M ctx =C2=B7 $0.43/$0.87 per M =C2=B7 the heavier end of the V4 family =C2= =B7 Apr 24 xiaomi/mimo-v2.5-pro ( https://openrouter.ai/xiaomi/mimo-v2.5-pro?utm_campa= ign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&u= tm_medium=3Demail_action&utm_source=3Dcustomer.io ) V2.5 line, successor to V2 =C2=B7 1M ctx =C2=B7 $1.00/$3.00 per M =C2=B7 Ap= r 22 openai/gpt-5.4-image-2 ( https://openrouter.ai/openai/gpt-5.4-image-2?utm_c= ampaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final= +B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) Image generation, a new modality on the API surface =C2=B7 April release 35 new models went live in April across Anthropic, DeepSeek, Moonshot, Open= AI, Xiaomi, Alibaba, xAI, and Google. See all April launches =E2=86=92 ( ht= tps://openrouter.ai/models?order=3Dnewest&utm_campaign=3D050726_monthly-rec= ap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action= &utm_source=3Dcustomer.io ) Platform in numbers Models 300+ Across 60+ providers Tokens / month 70T Processed on platform Developers 5M+ Globally on OpenRouter Browse all models ( https://openrouter.ai/models?utm_campaign=3D050726_mont= hly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail= _action&utm_source=3Dcustomer.io ) Latest from the blog GPT-5.5 price increase: what it actually costs ( https://openrouter.ai/anno= uncements/gpt55-cost-analysis?utm_campaign=3D050726_monthly-recap_s1&utm_co= ntent=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source= =3Dcustomer.io ) GPT-5.5 price increase: what it actually costs ( https://openrouter.ai/anno= uncements/gpt55-cost-analysis?utm_campaign=3D050726_monthly-recap_s1&utm_co= ntent=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source= =3Dcustomer.io ) OpenAI doubled GPT-5.5's per-token price, but the model writes less. We mea= sured real usage to find the net cost impact. Read the analysis =E2=86=92 ( https://openrouter.ai/announcements/gpt55-cos= t-analysis?utm_campaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+= Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) What Opus 4.7's tokenizer actually costs ( https://openrouter.ai/announceme= nts/opus-47-tokenizer-analysis?utm_campaign=3D050726_monthly-recap_s1&utm_c= ontent=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source= =3Dcustomer.io ) What Opus 4.7's tokenizer actually costs ( https://openrouter.ai/announceme= nts/opus-47-tokenizer-analysis?utm_campaign=3D050726_monthly-recap_s1&utm_c= ontent=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source= =3Dcustomer.io ) A million real requests analyzed. Up to 27% higher cost on most prompts, 1.= 6% lower under 2K tokens. Read the analysis =E2=86=92 ( https://openrouter.ai/announcements/opus-47-t= okenizer-analysis?utm_campaign=3D050726_monthly-recap_s1&utm_content=3DApri= l+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source=3Dcustomer.i= o ) Response caching: zero cost for identical requests ( https://openrouter.ai/= announcements/response-caching?utm_campaign=3D050726_monthly-recap_s1&utm_c= ontent=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source= =3Dcustomer.io ) Response caching: zero cost for identical requests ( https://openrouter.ai/= announcements/response-caching?utm_campaign=3D050726_monthly-recap_s1&utm_c= ontent=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source= =3Dcustomer.io ) A new Response Caching header. Identical API requests come back in a fracti= on of the latency, at zero cost. Read the post =E2=86=92 ( https://openrouter.ai/announcements/response-cach= ing?utm_campaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-= +V3+Final+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io ) Up next. May focuses on agent infrastructure: spend analytics by workspace,= per-key rate limits, and the next routing primitive. View announcements = =E2=86=92 ( https://openrouter.ai/announcements?utm_campaign=3D050726_month= ly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_= action&utm_source=3Dcustomer.io ) openrouter.ai ( https://openrouter.ai?utm_campaign=3D050726_monthly-recap_s= 1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm= _source=3Dcustomer.io )=C2=B7Discord ( https://discord.gg/openrouter?utm_ca= mpaign=3D050726_monthly-recap_s1&utm_content=3DApril+2026+Recap+-+V3+Final+= B&utm_medium=3Demail_action&utm_source=3Dcustomer.io )=C2=B7X ( https://x.c= om/openrouter?utm_campaign=3D050726_monthly-recap_s1&utm_content=3DApril+20= 26+Recap+-+V3+Final+B&utm_medium=3Demail_action&utm_source=3Dcustomer.io )= =C2=B7Docs ( https://openrouter.ai/docs?utm_campaign=3D050726_monthly-recap= _s1&utm_content=3DApril+2026+Recap+-+V3+Final+B&utm_medium=3Demail_action&u= tm_source=3Dcustomer.io ) 169 Madison Avenue, New York, NY 10016 Manage preferences ( https://track.customer.io/manage_subscription_preferen= ces/dgSE_QoDAIvZ9gKK2fYCAZ4DMOtv_Kglb503qIvl1Q=3D=3D )|Unsubscribe ( https:= //track.customer.io/unsubscribe/dgSE_QoDAIvZ9gKK2fYCAZ4DMOtv_Kglb503qIvl1Q= =3D=3D ) --8b5956134fd841ece819830beee882642d3faa0d3520dd712e092e51e95a Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset="utf-8" OpenRouter
Now generate video from text or images through one unifi= ed API. Use a TypeScript SDK to turn any model into an agent. Organize your= OpenRouter projects into separate environments with Workspaces. =E2=80=87= =CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2= =80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD= =8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80= =87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F = =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87= =CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2= =80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD= =8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80= =87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F = =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87= =CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =E2=80=87=CD=8F =C2=AD =C2=AD =C2=AD= =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2= =AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD = =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2= =AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD = =C2=AD =C2=AD =C2=AD =C2=AD =C2=AD=C2=A0
3D"OpenRouter" 3D"OpenRouter"

What shipped in April.

Now generate video from text or= images through one unified API. Use a TypeScript SDK to turn any model int= o an agent. Organize your OpenRouter projects into separate environments wi= th Workspaces. And 35 new models landed. Here's the recap.

Platform features
=20 =20 =20 =20
01
Video generation

Video APIs are fragmente= d. Providers use different request shapes, parameter names, and billing uni= ts. We built one API around those differences.

Submit a prompt to /api/v1/videos, get a job ID back, retri= eve the video when ready. Parameters are normalized across every model (res= olution, duration, aspect ratio, audio, frame images, reference images). A = capability-discovery endpoint tells you what each model supports before you= call it.

sora-2-pro veo-3.1 seedance-2.0 wan-2.7
Browse video models =E2=86=92|Try the demo =E2= =86=92
02
Workspaces

Separate environments fo= r separate projects, each with its own API keys, routing defaults, guardrai= ls, and observability dashboard. Useful for keeping side projects, prototyp= es, and production cleanly isolated under one account.

Create a workspace =E2=86=92<= span style=3D"display: inline-block; color: rgb(228, 228, 231); vertical-al= ign: bottom; margin: 10px 12px 0;">|Read the docs =E2=86=92
03
Agent SDK with callModel

One function handles mul= ti-turn tool calling, streaming, stop conditions, and cost tracking across = 300+ models. Stop conditions are composable: cap a session at Cost(1.00), stop when a specific tool gets cal= led, limit by turn count, or combine them. The @openrouter/agent TypeScript package ships with two scaffolds: = create-agent-tui for a terminal harness and create-headless-agent for a server-side one. Typed tool = context with Zod is supported.

SDK docs =E2=86=92|Build your own harness =E2=86=92
04
Benchmark scores on every model= page

Every model page now sho= ws benchmark scores from Artificial Analysis: coding, math, science, and lo= ng-context reasoning. Useful when deciding between models at similar price = points, or weighing open-weights against proprietary. The Rankings page picked up = the same data with scatter charts and long-context trends from 100K to 1M t= okens.

See the rankings =E2=86=92
05
I/O loggingBETA

The Activity log can now= capture the full request and response for every API call. Open any generat= ion to inspect the prompt, switch to Dev mode for raw JSON, or replay it in= chatroom against a different model.

Enable I/O logging =E2= =86=92
A= lso shipped in April

Reranker models with Cohere and Firework= s, Model Fusion for combining outputs across multiple models, prompt histor= y with replay and side-by-side comparison, Design Arena ELO rankings on the= comparison pages, knowledge cutoff dates exposed in the /models API, Stripe Projects for streamlined account = setup, and typed tool context with Zod validation in the TypeScript SDK.

New models
=20
Kimi K2.6
moonshotai/kimi-k2.6 =C2=B7 Apr 20

1.82T tokens since launch. 1T= -parameter MoE, 32B active per token. SWE-Bench Pro 58.6% beats GPT-5.4 and= Claude Opus 4.6. Open weights on Hugging Face.

$3.49/M out262K ctxopen weights
View on = /models =E2=86=92
=20
Claude Opus 4.7
anthropic/claude-opus-4.7 =C2=B7 Apr 16

935B tokens since launch. Ant= hropic's frontier reasoning and coding model. Beats Opus 4.6, GPT-5.4, and = Gemini 3.1 Pro across reasoning and code benchmarks.

$25/M out1M ctx
V= iew on /models =E2=86=92
=20
GPT-5.5 + GPT-5.5 Pro
openai/gpt-5.5 =C2=B7 Apr 24

OpenAI's latest. Pro variant = pushes harder on the reasoning ceiling at the cost of latency and price; th= e base model is the workhorse.

$30/M out1M ctxPro: $30/$180 per M
View on /models = =E2=86=92
deepseek/deepseek-v4-flash
704B tokens since launch =C2=B7= 1M ctx =C2=B7 $0.14/$0.28 per M =C2=B7 Apr 24
deepseek/deepseek-v4-pro
1M ctx =C2=B7 $0.43/$0.87 per M= =C2=B7 the heavier end of the V4 family =C2=B7 Apr 24
xiaomi/mimo-v2.5-pro
V2.5 line, successor to V2 =C2= =B7 1M ctx =C2=B7 $1.00/$3.00 per M =C2=B7 Apr 22
openai/gpt-5.4-image-2
Image generation, a new modalit= y on the API surface =C2=B7 April release

35 new models went live in April across = Anthropic, DeepSeek, Moonshot, OpenAI, Xiaomi, Alibaba, xAI, and Google. See all April launches =E2=86=92

Platform in numbers
Models
300+
Across 60+ providers
Tokens / month
70T
Processed on platform
Developers
5M+
Globally on OpenRouter
Browse all models=
Latest from the blog
3D"GPT-5.5 GPT-5.5 price increase: what it actually costs

OpenAI doubled GPT-5.5's per-tok= en price, but the model writes less. We measured real usage to find the net= cost impact.

Read the anal= ysis =E2=86=92
3D"What What Opus 4.7's tokenizer actually costs

A million real requests analyzed= =2E Up to 27% higher cost on most promp= ts, 1.6% lower under 2K tokens.

Rea= d the analysis =E2=86=92
3D"Response Response caching: zero cost for identical requests

A new Response Caching header. I= dentical API requests come back in a fraction of the latency, at zero cost.=

Read the post =E2= =86=92
Up next. May focuses on agent infrastructur= e: spend analytics by workspace, per-key rate limits, and the next routing = primitive. View anno= uncements =E2=86=92
openrouter.ai=C2= =B7Discord=C2=B7X=C2=B7Docs
169 Madison Avenue, New York, NY 10016
Manage preferences|Unsubscribe
3D""/ --8b5956134fd841ece819830beee882642d3faa0d3520dd712e092e51e95a--