ModelsNews

Claude Fable 5.2, Kimi K3.1 Pro and DeepSeek V4.1 Flash

Anthropic's newest Fable, Moonshot's Kimi K3.1 Pro and a faster DeepSeek V4.1 Flash are in the Chat model menu and on the AI Integrator runtime in Code.

Scroll down

Claude Fable 5.2 from Anthropic, Kimi K3.1 Pro from Moonshot AI and DeepSeek V4.1 Flash 1017 Ultrafast are in the Chat model menu. All three also run in Code on your AI Integrator plan, which now offers 43 models there. At current prices, all three need a Pro or Max plan.

What is Claude Fable 5.2?

Anthropic released Fable 5.2 on November 9 as its model for long, demanding work, such as agent sessions that run for hours or research carried through to a finished document. At launch, Anthropic listed it at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1, with a context window of 1 million tokens.[^1] Anthropic says early testers saw Fable 5.2 write its answers up to 30 percent faster than Fable 5.1.[^1]

On FrontierCode 2.0, a test of agentic coding, Anthropic reports 49.1 percent for Fable 5.2 with its effort setting at extra high, against 45.5 percent for Claude Opus 5.5 at its highest setting.[^1] These are Anthropic's own figures. Anthropic still recommends Opus 5.5 as the starting point for most coding and knowledge work, and Fable 5.2 when a task's size or difficulty calls for it.

Anthropic says it keeps Fable 5.2 requests for 30 days for safety monitoring, so AI Integrator can't offer it under Zero Data Retention.[^1] While Zero Data Retention is on in your privacy settings, which it is by default, Fable 5.2 appears in the menu but stays locked. Turning the setting off unlocks it, and also lets the companies that run your other models keep your requests under their own policies.

What is Kimi K3.1 Pro?

Moonshot AI calls Kimi K3.1 Pro the first model in its K3.1 series and its most capable open-weight model of its size so far. Open-weight means Moonshot publishes the model itself, so other companies can run it on their own servers. Moonshot built it for coding, 3D work, financial research and other tasks where the model works through many steps on its own, and gives it a context window of 1 million tokens.[^2]

On Terminal-Bench 4.0, a test of coding in a terminal, Moonshot reports 65.7 percent for Kimi K3.1 Pro and 66.4 percent for Claude Opus 5.5.[^2] These are Moonshot's own figures.

What is DeepSeek V4.1 Flash 1017 Ultrafast?

DeepSeek V4.1 Flash 1017 is the October 17 update of DeepSeek's Flash model. Vercel lists the Ultrafast version as the same model served by Fireworks for faster replies, through the Vercel AI Gateway that AI Integrator uses to reach model hosts. Vercel gives it a context window of 1,048,576 tokens and, when this was written, listed it at $2.40 per million input tokens and $9.60 per million output tokens.[^3]

Vercel describes DeepSeek V4.1 Flash as built for fast, efficient work with text and images. The Ultrafast version costs more for its speed. The standard V4.1 Flash stays in the menu, listed by Vercel at $0.022 per million input tokens and $0.30 per million output tokens, for work where a few more seconds don't matter.[^3]

How to use them in Chat

Pick any of the three from the model menu before you send a message. The model is chosen per message, so a conversation can move to Fable 5.2 for one hard question and back to a cheaper model afterwards, as Chat with multiple AI models shows. Each model reads everything said so far in the conversation.

At the gateway's prices when this was written, each costs more than $5 per million output tokens,[^4] so all three need Pro or Max. The app decides from the current price, so this can change if a price does, and on Free a model that needs Pro says so when you pick it. They draw on the same monthly usage as every other model.

How to use them in Code

The AI Integrator runtime in Code runs each model in the coding program its tools were built for. Fable 5.2 runs in Claude Code, and Kimi K3.1 Pro and DeepSeek V4.1 Flash 1017 Ultrafast run in OpenCode. The usage comes out of your plan, so you don't need a login with Anthropic, Moonshot or DeepSeek. As in Chat, the three need Pro or Max, and Fable 5.2 stays locked while Zero Data Retention is on. Code on your AI Integrator plan explains how the runtime works.

AI Integrator models in Code, by program43 models. Source: AI Integrator's Code model list
AI Integrator's Code model list
Fable 5.2 is one of the five models in Claude Code. Kimi K3.1 Pro and DeepSeek V4.1 Flash 1017 Ultrafast are two of the 31 in OpenCode.

If you use Code with your own Claude subscription, update Claude Code to see Fable 5.2 in its model list. Anthropic decides which Claude plans include it.