AI Coding Live News

On November 4 Together AI launched Together Compiler (TKCC), which pairs a swarm of AI agents with a persistent inference engine to keep rewriting inference kernels for real traffic, models and hardware, as part of what it calls Polymorphic Inference. The company says that within 24 hours it reached up to 2.6x higher TPS on models such as Kimi K3, GLM 5.3, MiniMax M3.1-Flash and DeepSeek v4.1-Flash, up to 50% faster on customer workloads, and it already serves trillions of tokens a day in production.

Mistral AI announced on X on Oct 28 that its Responses API is in public preview, positioned as the interface for building agents on Mistral. The official docs say it uses the OpenAI Responses format, and each response's output array carries the models messages, its reasoning and its function calls as separate items, which is easier to handle step by step than Chat Completions. The preview is stateless, so the app keeps the conversation history and sends it back; it supports text, image and file input, JSON Schema structured output, streaming events, client-executed function calls, reasoning.effort, and token counting via POST /v1/responses/input_tokens. It is served in the EU and US on Mistral AI Cloud Inference. The same thread says Mistral now also serves DeepSeek V4.1 Flash (public preview) and GLM 5.3 Flash (generally available).

CrowdStrike Intelligence found Claude Code session histories, ARTEX configs and a CLAUDE.md in attacker open directories, showing a likely Chinese-speaking, financially motivated actor used the China-built open-source agentic pentest framework ARTEX (DeepSeek v4.1-flash primary, plus GLM-5.3 and Grok 4.6) to breach loan-inquiry and employee mobile systems at South Korean financial institutions from late September to early October 2026.

Developer Taro L. Saito (@taroleo) followed a TypeSafe Jev-style approach—dropping full text/JSON generation and specializing on decisions—to distill DeepSeek V4 Flash judgments (~157GB weights) into a 4B model with about 26 hours of training on a DGX Spark. The student is roughly 1/20 the size, beats the teacher’s instant-answer mode, and runs about 22ms per decision, showing a fast local-LLM decision engine is practical.

boltdotnew announced another Bolt Forge addition after its open-model week: DeepSeek V4.1 Flash, with about 10x the usage of DeepSeek V4 Pro. The team says beta users report it builds the best-looking sites, pushing a faster open-weight option further into frontend generation workflows.

bolt.new introduced Bolt Forge, an open-weight coding agent now live in the bolt.new model picker. Free until Oct 14 with up to ~50x usage and zero usage charges; lineup includes GLM 5.3 / GLM 5.3 Flash, Kimi K3, and DeepSeek v4 Pro. Opt-in builds help train a trillion-parameter-class model with Arcee; also announced $9/mo Bolt Lite (Forge included; join by Oct 14 to lock $9). Partners include Arcee, Microsoft, Vercel, Fireworks, and DigitalOcean.

OpenRouter reported DeepSeek V4.1 Flash processed ~1T tokens in 24 hours and is on pace for ~2.8T in 48 hours—potentially its biggest paid-model launch window. About 90% of those tokens were cache reads, priced by the market at ~$0.006/M, roughly 5x cheaper than comparable models like GLM-5.3 Flash.

Cerebras said Alibaba's open-weight dense Qwen3.8-27B is live at Cerebras speed. It scores 34 on the Artificial Analysis Intelligence Index—comparable to models such as GPT-5.6 Luna, DeepSeek V4 Pro, and Claude Sonnet 4.6. Qwen also confirmed rapid inference on Cerebras.

Open-source Codex skill Fable Orchestrator (k2s1/Fable-Orechester-) routes one task across specialized models: GPT-5.6 Luna for implementation, DeepSeek V4 Flash for loops, and Claude Fable 5.1 for planning and final adjudication. After install, run /fable orchestrator [task].