Agents hit production, but trust is still being built.
Claude Opus 5, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
Anthropic
Claude Opus 5
DEPLOYif you run coding or knowledge work agents
WHAT SHIPPEDClaude Opus 5 is available today at half the price of Claude Fable 5. It more than doubles Opus 4.8 on Frontier-Bench v0.1 and scores 10.2 percentage points higher on organic chemistry tasks.
THE CATCHThe model is behind Mythos 5 on cybersecurity tasks, so security-sensitive workloads should not assume parity with the frontier.
IF YOU RUNa trading firm, an engineer built a market data feed for a new exchange in a single session
SOURCEAnthropic, Introducing Claude Opus 5
The Deploy Log
Models and APIs
| WATCHif you run enterprise customer support OpenAI Presence. OpenAI Presence is available today for voice and chat agents, and it resolves 75% of inbound issues without human assistance. OpenAI |
| WATCHif you run a small business ChatGPT for small business program. OpenAI is launching the ChatGPT for small business program with virtual training, in-person AI academies, and new guides, after 78% of participants built a functional AI workflow in a single day at last year's Small Business AI Jams. OpenAI |
Tools and products
| DEPLOYif you route models through AI Gateway Gemini 3.6 Flash. Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are now available on AI Gateway with no platform fee on inference. Vercel |
| WATCHif you run token-efficient agentic inference Ling 3.0 Flash. Ling 3.0 Flash from Ant Group is now available on AI Gateway and is free to use for the next three weeks, through August 3rd. Vercel |
| WATCHif you run agentic coding tasks Laguna S 2.1. Laguna S 2.1 from Poolside is now available on AI Gateway in free and paid versions, with the paid version supporting a 1M token context window. Vercel |
| DEPLOYif you optimize latency versus cost AI Gateway service tiers. AI Gateway now supports service tiering with default, priority, and flex tiers, where priority costs about 1.8 to 2 times default pricing and flex costs about 0.5 times default pricing. Vercel |
| DEPLOYif you run long workflow steps Workflow extended durations. Workflow steps on Pro and Enterprise plans can now run for up to 30 minutes, up from 800 seconds, using extended function durations in beta. Vercel |
| DEPLOYif you run Python functions on Vercel Python precompiled bytecode. Vercel now compiles Python functions to bytecode at build time, and cold starts for the median-sized function dropped from 2.8 seconds to 1.3 seconds in benchmarks. Vercel |
| DEPLOYif you integrate external services Vercel Connect preset connectors. Vercel Connect now includes preset connectors for over 100 services, including Shopify, Resend, Sanity, Okta, Workday, and Jira. Vercel |
Left out
- Gemini 3.5 Flash-Lite: Covered within the Gemini 3.6 Flash lead; no separate lead slot needed.
- Gemini 3.5 Flash Cyber: Limited-access pilot program only, not broadly available.
- GPT-5.6 Sol: Mentioned only as part of Health in ChatGPT, not a standalone release.
- Sandbox dashboard management: a figure in the draft could not be found on the publisher's own page
- Vercel MCP purchases: it did not carry a number that moved
- eve extensions: it did not carry a number that moved
- GitHub tools eve extension: it did not carry a number that moved
- Vercel Flags CLI history: it did not carry a number that moved
- Vercel Flags evaluation metrics: it did not carry a number that moved
- Notion Workers usage dashboard: it did not carry a number that moved
- AI Gateway streaming transcription: it did not carry a number that moved
- Python WebSocket support: it did not carry a number that moved
- Vercel WAF for Blob: it did not carry a number that moved
- Vercel MCP deploys code: it did not carry a number that moved
- Vercel Agent: it did not carry a number that moved
- Anthropic Economic Index connector: it did not carry a number that moved
- Health in ChatGPT: the condition it attaches to the call did not hold up
The pattern
The week's releases collapse the distance between a model card and a production workflow. Claude Opus 5 ships at half the price of Claude Fable 5 and more than doubles Opus 4.8 on Frontier-Bench v0.1, while OpenAI Presence resolves 75% of inbound issues without human assistance and the ChatGPT for small business program follows 78% of participants building a functional AI workflow in a single day. Vercel turns that capability into infrastructure: Gemini 3.6 Flash and Gemini 3.5 Flash-Lite arrive on AI Gateway with no platform fee on inference, Ling 3.0 Flash is free through August 3rd, and Laguna S 2.1 offers a 1M token context window in a paid version. Service tiers price priority at 1.8 to 2 times default and flex at 0.5 times default, while Workflow steps on Pro and Enterprise plans extend to 30 minutes from 800 seconds and Python cold starts drop from 2.8 seconds to 1.3 seconds.
But the same items show trust is still being built, not assumed. Claude Opus 5 is behind Mythos 5 on cybersecurity tasks, so security-sensitive workloads should not assume parity with the frontier. Vercel Connect preset connectors for over 100 services, including Shopify, Resend, Sanity, Okta, Workday, and Jira, make integration the default rather than the exception, yet the catch on Opus 5 is the only explicit warning in the set. A trading firm built a market data feed for a new exchange in a single session, which says more about the speed of agentic output than about its safety in every domain.