Claude Opus 5 expands on Vercel, AWS, Replicate, and OpenRouter plus fresh Hugging Face models

What shipped

On 24 July 2026 several vendors released updates that move AI from chat interfaces toward longer-running agents and production workflows. Claude Opus 5 received the widest distribution, while Hugging Face surfaced new open models for coding, vision, and African languages. The day also brought platform changes at Vercel and Replicate plus one acquisition in the agent space.

Hugging Face trending

Hugging Face hosted the largest set of new model releases with six entries spanning text generation, vision, speech recognition, and agent training. Poolside and Kwaipilot led the text models while baseten added a vision variant. Three research papers also appeared on agent harnesses, African-language ASR, and reliability for non-engineer-built agents.

• •Laguna-S-2.1-NVFP4 Poolside released a text-generation model that runs on vLLM and supports fine-tuning on the Hub for coding and chat tasks.
• •KAT-Coder-V2.5-Dev Kwaipilot released a transformers-based text model aimed at developer workflows and available for direct inference or further training.
• •GLM-5.2-Vision-NVFP4 Baseten released an image-text-to-text model built on sglang that handles document and diagram understanding.
• •OpenForgeRL A new training harness lets teams run reinforcement learning on agents inside open environments instead of closed inference stacks.
• •DONDO Twenty-one monolingual and five multilingual ASR models cover twenty-seven African language varieties using the w2v-BERT 2.0 encoder.
• •Continuous assurance paper Research outlines reliability checks for agents that non-engineers build inside low-code tools used by small teams.

Vendor launches

Vercel shipped the largest set of practical changes with longer workflow steps and Claude Opus 5 on its AI Gateway. Anthropic's model also landed on AWS and OpenRouter. Two policy and partnership notes from Google and NVIDIA rounded out the vendor activity.

• •Vercel extended durations Workflow steps on Pro and Enterprise plans now run up to 30 minutes after an environment variable toggle, useful for long agent tasks.
• •Claude Opus 5 on AI Gateway Vercel added Anthropic's model for multi-file refactors and end-to-end feature work at lower token counts than prior Opus versions.

Replicate new models

Replicate added two distinct models after removing duplicates. Anthropic placed its newest Sonnet variant for agentic coding while a separate image model supports seamless pattern generation.

• •claude-sonnet-5 Anthropic released its most agentic Sonnet model on Replicate for frontier coding and tool use at Sonnet speed and price.
• •mindprint-flux2-klein Okgodoit-repos released a FLUX.2-based generator that creates seamless repeating patterns for design and texture work.

Product Hunt picks

Buzz: The tool places people, agents, and project data together so small teams can manage work without switching between multiple apps.

Other

AWS and OpenRouter both added Claude Opus 5 while Notion acquired an AI search startup. AllenAI published a post on open models and one duplicate OpenRouter listing was removed.

• •Claude Opus 5 on AWS Anthropic's flagship model is now available for demanding reasoning and long-horizon coding tasks inside AWS environments.
• •ZeroEntropy joins Notion Notion acquired the AI search company to strengthen document retrieval inside its workspace product.
• •AllenAI open models post The institute argues that fully open weights and research are required for independent review and continued scientific progress.
• •Claude Opus 5 Fast on OpenRouter A higher-speed variant offers the same capabilities at double the regular Opus 5 price for time-sensitive jobs.
• •Claude Opus 5 on OpenRouter The standard version provides 1,000k context at $5 per million input tokens for large-scale agent runs.

Industry news

Anthropic's Opus 5 launch dominated coverage with claims of near-Fable 5 performance at half the cost. Cognition acquired Poke to improve agent personality while Microsoft pushed open-weight models for Azure. One unrelated medical story was removed.

• •Anthropic Opus 5 launch The new model scores near Fable 5 on coding benchmarks while charging roughly half the token price of the competitor.
• •Microsoft open-weight letter Microsoft joined Meta and NVIDIA in calling for open weights to increase Azure model usage and reduce OpenAI spend.
• •Cognition acquires Poke The Devin maker bought the conversational AI to add friend-like interaction style to its coding agent.
• •ARC-AGI-3 scores Opus 5 reached 30.2 percent on the novel-problem benchmark, nearly four times higher than GPT-5.6 Sol.
• •Open-weight Azure strategy Analysts note the move lets Microsoft swap in cheaper in-house models and cut reliance on external frontier APIs.

What this means for you

For Vibe Builders: You can now call Claude Opus 5 directly from Vercel, Replicate, or OpenRouter to run longer agent workflows without managing servers. New Hugging Face models for vision and speech give quick starting points for side projects, and the Buzz tool on Product Hunt shows how teams are bundling agents with project data in one place. Test the 30-minute Vercel steps on a real multi-file task this week.

For Non-techies: Claude Opus 5 is now easier to reach through familiar platforms like Vercel and AWS, so small businesses can hand off bigger coding or document tasks without new accounts. Notion's acquisition of ZeroEntropy hints that search inside workspaces will keep improving. Watch for the longer workflow runs on paid Vercel plans if you already use their hosting.

For Developers: Opus 5 on multiple gateways plus the new Sonnet variant on Replicate let you benchmark cost and latency across providers before committing. Hugging Face releases include training harnesses and African-language ASR that can slot into existing pipelines. Check the ARC-AGI-3 numbers and the 30-minute Vercel limit against your current agent runtimes before the next integration cycle.

What to watch next

Track how teams adopt the 30-minute Vercel steps and whether Opus 5 Fast sees uptake on OpenRouter. Watch for follow-up papers on the agent assurance topic and any new benchmarks comparing Opus 5 to Fable 5 on production coding tasks.

Sources

Hugging Face trending

• •Laguna-S-2.1-NVFP4 by poolside trends on HuggingFace
• •KAT-Coder-V2.5-Dev by Kwaipilot trends on HuggingFace
• •GLM-5.2-Vision-NVFP4 by baseten trends on HuggingFace
• •OpenForgeRL: Train Harness-native Agents in Any Environment
• •DONDO: Open w2v-BERT Speech-Recognition Base Models for African Languages
• •Toward Continuous Assurance for the Democratization of AI Agent Creation in Industry

Vendor launches

• •Workflow steps now support extended function durations
• •Claude Opus 5 now available on AI Gateway
• •Google is signing the EU AI Act Code of Practice on Transparency of AI-Generated Content.
• •At AI Summit, South Korea Outlines Its AI Future With NVIDIA and Partners

Replicate new models

• •claude-sonnet-5 by anthropic launches on Replicate
• •mindprint-flux2-klein by okgodoit-repos launches on Replicate

Product Hunt picks

• •Buzz

Other

• •Introducing Claude Opus 5 on AWS: Anthropic’s most capable Opus model
• •ZeroEntropy is joining Notion
• •2022 IEEE President K.J. Ray Liu Honored for His Leadership
• •Who gets to understand AI?
• •Claude Opus 5 (Fast) now available on OpenRouter (1,000k context, $10.00/M in, $50.00/M out)
• •Claude Opus 5 now available on OpenRouter (1,000k context, $5.00/M in, $25.00/M out)

Industry news

• •Anthropic launches Opus 5
• •The quest to keep organs alive outside the body
• •Did Chinese AI Steal From Anthropic, and OpenAI Loses Control of Two Models
• •Microsoft's open-weight AI push is so obviously an Azure play it hurts
• •Why Cognition bought Poke: AI personality is becoming a competitive advantage
• •Anthropic claims its new Claude Opus 5 delivers near-Fable 5 performance at half the token price