DeepSeek-V4-Flash

2026-10-26

RESEARCH · 1 SOURCE · arXiv cs.AI

HyScale paper introduces HyTopK and selection replay to speed long-context learned sparse attention training

HyScale (arXiv:2610.32057v1) presents a system for scaling long-context training of learned sparse attention by treating selection outputs as index sets and introducing HyTopK (on-the-fly exact top-k) plus selection replay to avoid full score tables and repeated selection. The paper reports up to 2.07× speedup training GLM-5.1 on 1M-token sequences versus Megatron-LM on 512 H20 GPUs, and 2.01× / 1.57× speedups for the CSA layers of DeepSeek-V4-Flash and the QSA layers of Qwen3.8-Flash-Next at 1M tokens on 32 H800 GPUs.

7.0

2026-10-16

RESEARCH · 1 SOURCE · arXiv cs.AI

IneqAgent: LLM-guided agent for inequality proving in Lean4

IneqAgent is an agentic framework that combines LLM reasoning with symbolic computation to prove inequalities in Lean4. It uses lemma-level Monte Carlo Tree Search to coordinate lemma construction, goal reformulation, and selection among exact-symbolic backends (sum-of-squares, Bernstein, successive difference-substitution); on four benchmarks with DeepSeek-V4-Flash it solves 175 of 186 problems (94.09%) and produces Lean-verified strengthenings of Dittert's conjecture (dimensions 4 and 5) and a certification of the Holens--Djokovic (4,4) inequality.

7.0

2026-09-21

RESEARCH · 1 SOURCE · arXiv cs.AI

GameASG-Bench: benchmark for behavioral testability in autonomous game software generation (arXiv:2609.21293v1)

The paper introduces GameASG-Bench, a benchmark that requires behavioral testability for autonomous software generation in game development by specifying an evaluation interface and providing static L1 source checks plus browser-executed L2 runtime checks. The release contains 47 browser-native game-generation tasks across 12 genres with reference implementations; experiments on nine agent stacks report up to 93.2% mean L2 check pass rate but only a 55.3% strict task success rate (26/47), and show gains for DeepSeek-V4-Flash with full tool access and larger nominal turn budgets.

7.0

2026-09-18

RESEARCH · 1 SOURCE · arXiv cs.AI

little m: an AI agent to formulate industrial process optimization models

Researchers released 'little m', an AI agent that combines a domain-specific knowledge repository with LLM-driven interaction to translate messy, multimodal industrial specifications (text and process diagrams) into mathematical optimization models. They also introduced IPC-Bench, a 50-scenario multimodal benchmark for industrial process control; evaluations (automated structural checks and double-blind human review) show little m generates substantially more semantically correct formulations than state-of-the-art LLMs, though the paper does not evaluate solver feasibility, physical validity, or closed-loop performance.

6.0

2026-09-10

MODELS · 1 SOURCE · DeepSeek

DeepSeek launches DeepSeek‑V4.1‑Flash with new causal encoder–decoder and lower costs

DeepSeek announced DeepSeek‑V4.1‑Flash, a smaller-native-vision model using a new Causal Encoder–Decoder design (8B active params for input, 16B for output), new pretraining and larger-scale RL post-training, plus KV‑cache compression. The company is retiring V4‑Flash variants, routing deepseek-v4-pro traffic to V4.1‑Flash from 04:00 UTC on Sept 14, 2026, and changing pricing (new rates take effect 04:00 UTC on Sept 10, 2026) while partners WorkBuddy, CodeBuddy and OpenCode already support V4.1‑Flash.

8.0

2026-04-24

MODELS · 1 SOURCE · DeepSeek

DeepSeek publishes open-source DeepSeek‑V4 preview with 1M-token context

DeepSeek has open-sourced a preview of DeepSeek‑V4, offering two variants: DeepSeek‑V4‑Pro (1.6T total / 49B active params) and DeepSeek‑V4‑Flash (284B total / 13B active params). The company says both models support a default 1M-token context, introduce novel token-wise compression and DSA (DeepSeek Sparse Attention), claim open-source state‑of‑the‑art agentic and reasoning performance (trailing only Gemini‑3.1‑Pro among models they compare to), and are available now via chat.deepseek.com and updated APIs with compatibility for OpenAI ChatCompletions & Anthropic endpoints; older models deepseek-chat and deepseek-reasoner will be retired on Jul 24, 2026 at 15:59 UTC.

8.0