Frontier models get cheaper while the tooling around them gets more granular
Claude Opus 4.5, FLUX.2 and FLUX.2 Pro on AI Gateway, 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 4.5
DEPLOYif you run long-horizon coding agents
WHAT SHIPPEDClaude Opus 4.5 is available today on Anthropic's apps, API, and all three major cloud platforms at $5 per million input tokens and $25 per million output tokens. It leads 7 out of 8 programming languages on SWE-bench Multilingual and scores a 10.6% jump over Sonnet 4.5 on Aider Polyglot.
THE CATCHThe model scored higher than any human candidate ever on Anthropic's internal performance engineering take-home exam within the 2-hour time limit, and it found a legitimate workaround on τ2-bench that the benchmark scored as a failure because the solution was unanticipated.
IF YOU RUNproduction code review at scale, Opus 4.5 catches more issues without sacrificing precision
SOURCEAnthropic, Introducing Claude Opus 4.5
Hugging Face
FLUX.2
WATCHif you generate images on 24GB GPUs
WHAT SHIPPEDFLUX.2 is a new open image generation model from Black Forest Labs with a 32B parameter DiT and a single Mistral Small 3.1 text encoder. It supports multiple reference images and runs on 24GB GPUs with 4-bit quantization.
THE CATCHFLUX.2 is not a drop-in replacement for FLUX.1. Unquantized inference takes more than 80GB VRAM, and the architecture shifts from 19 double-stream to 38 single-stream blocks in Flux.1 to 8 double-stream to 48 single-stream blocks.
IF YOU RUNimage generation on a single consumer GPU, the 4-bit path needs about 20GB of free VRAM
SOURCEHugging Face, FLUX.2 - BFL's new open image generation model
Vercel
FLUX.2 Pro on AI Gateway
DEPLOYif you already route through Vercel AI Gateway
WHAT SHIPPEDFLUX.2 Pro from Black Forest Labs is now available via Vercel's AI Gateway with no other provider accounts required. Set model to bfl/flux-2-pro in the AI SDK to generate images up to 4MP with multi-reference input.
THE CATCHFLUX.2 Pro is a pure image-focused rectified-flow transformer model, not a multimodal LLM like the other image-capable models already in AI Gateway.
IF YOU RUNimage generation through a unified API with retries and failover
SOURCEVercel, FLUX.2 Pro image model is now available on Vercel AI Gateway
The Deploy Log
Models and APIs
| WATCHif you compare products across many retailer sites ChatGPT shopping research. Shopping research in ChatGPT is rolling out today on mobile and web for logged-in users on Free, Go, Plus, and Pro plans, powered by a version of GPT-5 mini trained with reinforcement learning specifically for shopping tasks. OpenAI |
| WATCHif you need EU data sovereignty for inference OVHcloud Inference Provider. OVHcloud is now a supported Inference Provider on the Hugging Face Hub, offering serverless access to open-weight models like gpt-oss, Qwen3, DeepSeek R1, and Llama with pay-per-token pricing starting at €0.04 per million tokens. Hugging Face |
Tools and products
| DEPLOYif you need image generation without multimodal LLM Image-only models on AI Gateway. Vercel AI Gateway now offers image-only models including Black Forest Labs FLUX.2 Flex, FLUX.2 Pro, FLUX.1 Kontext Max, FLUX.1 Kontext Pro, FLUX 1.1 Pro Ultra, FLUX 1.1 Pro, FLUX.1 Fill Pro, and Google Imagen 4.0 Generate, Fast Generate, and Ultra Generate. Vercel |
| WATCHif you render markdown in React apps Streamdown 1.6. Vercel Streamdown 1.6 is now available with memoization, LRU caching, optimized string operations, removal of regexes, lazy-loaded Code Blocks, Mermaid, and Math components, a rebuilt code highlighting system, a custom markdown renderer, and Static Mode. Vercel |
| WATCHif you use the Vercel CLI daily vercel open CLI command. You can now open your current project in the Vercel Dashboard directly from the command line using vercel open, available in vercel 48.10.0 or newer. Vercel |
Left out
- Data residency expansion: a figure in the draft could not be found on the publisher's own page
- Node.js 24 LTS: a figure in the draft could not be found on the publisher's own page
- Advanced sampling rules for Drains: it did not carry a number that moved
- Convex on Vercel Marketplace: it did not carry a number that moved
- Cryptographically-verified commits: it did not carry a number that moved
- New TLDs on Vercel Domains: it did not carry a number that moved
- Sign in with Vercel: it did not carry a number that moved
- Claude Opus 4.5 on AI Gateway: its own name for the thing was too long to print
The pattern
The week's biggest model releases, Claude Opus 4.5 and FLUX.2, both arrive with a hard constraint attached: Opus 4.5 drops to $5 per million input tokens to make frontier coding affordable, while FLUX.2 demands 80GB VRAM unquantized and only becomes practical on 24GB GPUs through 4-bit quantization.
The tooling layer is consolidating around gateways and dashboards. Vercel added both new models to AI Gateway within days of release, and the rest of the log is granular platform work: commit verification, sampling rules, sign-in methods, and CLI shortcuts that tighten the deployment path rather than expand the model frontier.