🎉 Muse Spark 1.3 and Jev 1.13 are live — one chat model, and one API that only answers closed questions

Meta's muse-spark-1.3 is open on the spark-premium group and uses Chat Completions. The official token price is input $1.25, output $4.25, and cache read $0.15 per 1M tokens. The group multiplier is 1.15, so this group is input $1.4375, output $4.8875, and cache read $0.1725. Select spark-premium on the API key and set the model to muse-spark-1.3. Live prices: muse-spark-1.3.

TypeSafe's jev-1.13 is open on the jev-premium group. It does not write free text, so do not send it to the chat API. The body is a state plus closed questions of type noul, choice, or score. Each question comes back as a probability. Call POST /v1/decisions with the model jev-1.13. The official price is input $0.042 per 1M tokens and output $0. At the same 1.15 multiplier, this group is input $0.0483 and output $0. Live prices: jev-1.13. Billing follows the pricing page.

The blog has an analysis and breakdown of Jev, starting at JEV - A joint does not need a brain. There are ten pieces.

🎉 Grok 4.7 is live — Extra High beats same-tier Opus 5 on CursorBench

xAI's new flagship grok-4.7 is now available on Modelflare. It has a 500K context window, accepts text and image input, and returns text. Chat Completions and the Responses API are both supported. grok-4.5 and grok-4.6 stay available.

On CursorBench, Grok 4.7 Extra High scores 46.3%, ahead of Opus 5 Extra High at 46.1%. Cursor's published-price cost is about $6.01 per task, compared with $11.43 for Opus 5 Extra High. Opus 5 Max scores 46.6%. Cursor notes that scores vary, and small gaps may not be statistically meaningful. To match this effort level, set reasoning_effort to xhigh (the default is high).

Both groups are open. The list price matches grok-4.6, then the group multiplier applies:

• grok-award (0.03x): input $0.06, output $0.18, cache read $0.015 per 1M tokens
• grok-stable (0.11x): input $0.22, output $0.66, cache read $0.055 per 1M tokens

Pick the group on your API key and set the model to grok-4.7. Live prices: grok-4.7. Billing follows the pricing page.

🎉 Seedance 2.0 Fast and Mini are live — two cheaper Seedance 2.0 SKUs

ByteDance Seedance 2.0 now includes seedance-2.0-fast and seedance-2.0-mini. They use the same POST /v1/videos endpoint, at 480p / 720p and 4–15 seconds (or -1 to let the model pick the length). Text, image, video/audio references, and synced audio are supported.

Billing is token-based: Fast is 65% of the official list price, Mini is 52%. Jobs with video input use a lower unit price. Live quotes: seedance-2.0-fast and seedance-2.0-mini.

Use the Media group media on your API key, or generate in the Castaly workbench. Guides: seedance-2.0-fast and seedance-2.0-mini.

For 1080p / 4K or longer clips, keep using seedance-2.0 or seedance-2.5.

🎉 Castaly is live — image and video generation gets a studio of its own

Our image and video site, Castaly, is now live at castaly.modelflare.dev. Sign in with the Modelflare account you already have and start right away; there is no separate registration.

Create in the browser. Pick a model, write the prompt, adjust the parameters and review the result in one workspace. Everything you generate stays in your library, ready to revisit, download or delete.

Or call it locally. If you would rather work in a terminal, head to Castaly's Skill page: it gives you the install for the agent and operating system you actually use, and hands you the API key the calls need. Once it is in place, Claude Code, Codex and other Skill-aware tools can generate for you directly.

Your balance is shared. Castaly spends the same paid balance as the main site, so there is no second top-up, and usage and billing stay in one place (gift credit is still counted per product).

The platform carries 39 image and video models today: Seedream, GPT Image, Gemini Image and Wan Image on the image side, and Wan, Seedance, Kling, MiniMax and HappyHorse on the video side, 12 of them NSFW editions. Each model's request contract, parameter table and examples live on the docs site at docs.modelflare.dev, and live prices are on Models & Pricing.

🎉 Claude top-up group is live — any single top-up of US$50+ unlocks 0.03x pricing

OpenAI's official capacity has been tight lately and its routes keep running into problems. If your workload allows, we encourage you to use Claude more: the new claude-first-topup group is live, and from now on any single top-up of US$50 or more permanently unlocks Claude models at 0.03x (about 3% of the official price), covering claude-opus-5, claude-opus-4-8, claude-sonnet-5, claude-haiku-4-5, claude-fable-5 and claude-fable-5-1.

To use it, create or edit an API key under "API Keys" and pick the claude-first-topup group; smart keys include it automatically. The GPT top-up group (US$20+ for 0.015x) is unchanged, and a US$50+ top-up unlocks both.

Live prices are on Models & Pricing; top up from your Wallet.