Observed Signal · Sep 12, 2026 · Technical Release · Source: AINews swyx · Impact: 5/5 · Sentiment: Positive
DeepSeek Launches V4.1 Flash with Novel Encoder-Decoder Architecture
DeepSeek released DeepSeek-V4.1-Flash, a 763B-parameter mixture-of-experts model employing a novel causal encoder-decoder architecture with 8B active parameters for prefill and 16B for decode. It features native vision understanding, 1M token context, an MIT license, and extreme inference efficiency, claiming up to 1/8 KV cache footprint versus V4 Flash. Independent evals (Artificial Analysis Index 40, Vals Index #1 open-weight) show it surpasses V4 Pro at lower cost. API pricing is $0.30/1M input and $1.20/1M output tokens. DeepSeek has soft-retired V4 Pro, routing traffic to V4.1 Flash. The model supports SSD offload and local deployment, with Ollama and Baseten offering day-0 support. Technical discussions highlight the architecture's novelty and potential impact on long-context agents.
Technical release from DeepSeek, a major AI platform, introducing a revolutionary architecture with industry-shifting implications for inference cost and efficiency. Sets new benchmarks for open-weight models and could force competitive responses.
Track DeepSeek Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your Polaris7 workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- DeepSeek launched V4.1-Flash with a causal encoder-decoder architecture, 763B total params (8B prefill/16B decode active).
- Artificial Analysis Index scores V4.1-Flash at 40, above V4 Pro and below GLM-5.3-Flash.
- API pricing: $0.30 per 1M input tokens, $1.20 per 1M output tokens, cached input $0.006 per 1M.
- V4.1-Flash is #1 open-weight model on Vals Index, costing $0.30 per test.
- DeepSeek soft-retired V4 Pro, routing traffic to V4.1 Flash at Flash pricing.
- Independent eval: AutomationBench-AA 69%, GDPval-AA v2 Elo 1632 (surpassing Kimi K3), AA-LCR 84%.
- Local deployment reports: 200+ TPS on 4 Max-Qs with NVMe offload; 300+ TPS on 4 RTX Pros.
- Baseten and Ollama provide day-0 support for V4.1-Flash.
Connected Companies & Entities
4 Entities mapped“DeepSeek launched V4.1-Flash, its new open-weight flagship.”
“Independent benchmark account Artificial Analysis reported scores and pricing for V4.1-Flash.”
“Ollama began rolling out V4.1-Flash to Max and Team accounts.”
“Baseten shipped day-0 support for DeepSeek V4.1-Flash.”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
DeepSeek V4.1 Flash Beats Larger Model, Proving Ilya Right
Ilya Sutskever left OpenAI to found SSI and predicted the end of scaling. DeepSeek's new V4.1 Flash model, with a third of the parameters of its flagship V4-Pro, outperforms it on coding benchmarks (74.2 vs 62.7) while activating fewer parameters per token. Starting September 14, all V4-Pro requests route to V4.1 Flash. This validates the shift from model size to efficiency, marking a turning point in AI development.
DeepSeek V4 Flash Sets New 'Kill Line' in LLMs
DeepSeek released V4 Flash-0731, a 284B-parameter model shipped on July 31 that early benchmarks place above GLM-5.2 and competitive with Anthropic's Opus 4.8. The article introduces the "kill line" concept: models that are both more expensive and lower-performing than DeepSeek V4 are effectively uncompetitive. The release, combined with aggressive pricing and publicly available weights, pressures mid-tier proprietary models to either cut prices, run expensive modes, or become obsolete. DeepSeek plans a V4 Pro with roughly five times the parameters and a Harness agent framework in the coming weeks. The piece notes open-source adoption and a Chinese-language ecosystem barrier for Western developers, and warns that the kill line will shift as DeepSeek ships further dated releases every 2–3 months.
DeepSeek previews V4 open-source LLM
Deepseek on April 24, 2026 published its long‑anticipated Deepseek V4 (variants Pro and Flash), an open‑source large language model built on a new architecture with 1.6 trillion parameters. The company highlights significant gains in reasoning and autonomous code generation, claims benchmark-leading performance in mathematics, STEM and programming among open models, and says V4 supports context windows up to one million tokens while reducing compute and memory costs. Deepseek positions V4 Pro as materially cheaper on coding tasks versus OpenAI’s GPT‑5.5. The rollout also involves a partnership with Huawei, which supplies "Supernode" clusters of Ascend‑950 chips; Deepseek and analysts note a strategic focus on Huawei and Cambricon domestic chips to relieve reliance on Nvidia/AMD. Market reaction is expected to be more muted than Deepseek’s earlier 2025 breakthrough R1 shock.
Receive all future signals for DeepSeek & competitors in your briefing
Never miss material M&A deals, leadership changes, account wins, or product rollouts. Polaris7 monitors 20,000+ primary sources 24/7, delivering verified telemetry directly into your executive briefing.