B.AI reaches 2T token throughput in seven days offering free access to MiMo-V2.5, DeepSeek-V4-Flash and Tencent Hy3 models.

B.AI adds DeepSeek-V4-Flash and Tencent Hy3 with free access, pushing toward daily trillion-token throughput as TRON ecosystem gains traction at $0.34.

TRON founder Justin Sun highlights B.AI free DeepSeek-V4-Flash amid rival price hikes, with API access and 130M credits for new users.

B.AI hits 76.91B tokens and 2.06M users in 24 hours as Justin Sun highlights free DeepSeek access amid price hikes.

DeepSeek develops custom inference chip after crashing Nvidia $600 billion market cap, raising $7 billion amid US export controls and Huawei Ascend competition.

VitalikButerin details Deepseek v4 2-bit quant at 35 tok/s on Apple and CROPS AI Ethereum intersections while ETH trades at $2063.92.

DeepSeek cuts inference prices 75% permanently, reaching 50x lower costs than OpenAI and Anthropic with fractions of a cent per million tokens.

DeepSeek V4 Pro dominates EasyRouter platform with highest call volume, costing ¥84 for 600M tokens vs. half-year on 4090, highlighting AI model cost efficiency.

DeepSeek reduces input cache hit prices to 1/10th across all APIs, boosting AI model efficiency amid DeepSeek-V4-Pro 75% off promotion for developers.

DeepSeek cuts V4-Pro API costs by 75% until May 5, 2026, with updates for 1M context in Claude Code and more AI integration tools.

DeepSeek faces 2% odds of claiming the #1 AI model spot by June 30, per Polymarket predictions amid fierce AI industry competition.

DeepSeek rolls out deepseek-v4-pro and deepseek-v4-flash APIs today, supporting 1M context and dual modes amid AI model updates and large language model advancements.

According to Polymarket, there is a 9% chance that DeepSeek will have the top-ranked AI model by the end of June. This prediction highlights the competitive landscape in the AI sector, where achieving a leading position can significantly impact market dynamics and investor sentiment. Traders should monitor developments closely, as shifts in AI rankings could influence related stocks and technologies.

According to @KobeissiLetter, The Information reports that China’s DeepSeek is set to unveil a new AI model built using Nvidia’s Blackwell chips that the report characterizes as “smuggled” into China by dismantling data-center installations abroad and shipping them domestically (source: @KobeissiLetter citing The Information). The presence of Blackwell-class GPUs in China would run counter to U.S. Department of Commerce BIS restrictions on advanced AI chips to China announced in October 2023 and updated in October 2024 (source: U.S. Department of Commerce, Bureau of Industry and Security). For trading exposure, Nvidia (NVDA) is directly implicated as the chip supplier and has disclosed in SEC filings that export controls to China are a material risk (sources: NVIDIA investor relations and SEC filings). U.S. semiconductor baskets such as SMH concentrate large-cap AI chip suppliers and are often used for directional exposure to the theme (source: VanEck SMH fund materials). In crypto, AI-linked tokens that reference compute and AI, such as RNDR and FET, are listed by major exchanges under the AI sector, making them relevant tickers for AI-chip headlines (source: Binance exchange asset categories).

According to the source, Mistral announced a Frontier AI model family positioned directly against DeepSeek, indicating intensifying competition in cutting-edge model performance and deployment. According to the source, the post highlights a head-to-head framing but does not provide technical benchmarks, pricing, or release modalities, which are key variables for traders assessing impact on AI-linked crypto narratives. According to the source, traders should watch for confirmations on open-weights availability, API/inference pricing, and benchmark scores, as these details could shift sentiment and liquidity in AI infrastructure and compute-related crypto assets.

According to @casey_lau, Hangzhou is fast becoming China’s AI capital as a cohort dubbed the Six Little Dragons emerges as leaders, providing a clear local AI exposure map for traders; source: @casey_lau. The lineup spans DeepSeek (open-source LLMs), Unitree and DeepRobotics (robotics AI), BrainCo (brain-computer interface), ManyCore (spatial intelligence), and Game Science (Black Myth: Wukong), highlighting key verticals across models, robotics, BCI, spatial computing, and gaming for sector tracking; source: @casey_lau.

According to @scottshics, he has begun a thread to share the backstory of a whitepaper focused on agentic payments and AI agents transacting. Source: x.com/scottshics/status/1987536798269260002; x.com/GoKiteAI/status/1984254607069999116 He states the catalyst was early February 2025 after being shocked by DeepSeek’s reasoning performance, leading him to conclude that AI agents will act and transact and that existing infrastructure is not ready. Source: x.com/scottshics/status/1987536798269260002; x.com/GoKiteAI/status/1984254607069999116 He reports initial pushback from investors and his team, noting many saw payments as a solved problem and AI agents as science fiction. Source: x.com/scottshics/status/1987536798269260002; x.com/GoKiteAI/status/1984254607069999116 He situates this work before milestones he lists as A2A in April, x402 in May, and a Payment hype phase from August onward in 2025, framing a clear timeline for the AI-agent payment narrative. Source: x.com/scottshics/status/1987536798269260002; x.com/GoKiteAI/status/1984254607069999116 He confirms there were multiple internal debates on whether agentic payments were the right direction. Source: x.com/scottshics/status/1987536798269260002; x.com/GoKiteAI/status/1984254607069999116 He does not name any specific cryptocurrencies, tokens, or platforms in this update. Source: x.com/scottshics/status/1987536798269260002; x.com/GoKiteAI/status/1984254607069999116 The dated sequence of A2A (April), x402 (May), and Payment hype (August onward) provides reference points traders can use to map the 2025 AI-payment narrative. Source: x.com/scottshics/status/1987536798269260002; x.com/GoKiteAI/status/1984254607069999116

According to @PANewsCN, the Alpha Arena perpetual futures contest by nof1 put six leading AI models into a live $10,000 account challenge for 10 days, fully offline and trading solely via technical indicators. Source: PANews on X, Oct 28, 2025 https://twitter.com/PANewsCN/status/1983127032595898682 DeepSeek led with a 130% return to a $23,000 balance on just 17 trades, averaging 49 hours per position, a 41% win rate, and a 6.7:1 profit-loss ratio, driven by low-frequency trend following and 3-minute data reviews. Source: PANews on X, Oct 28, 2025 https://twitter.com/PANewsCN/status/1983127032595898682 Qwen3 posted a 100% return with per-trade leverage up to 5.6x and a maximum single loss exceeding $2,200, reflecting aggressive positioning with weak drawdown tolerance. Source: PANews on X, Oct 28, 2025 https://twitter.com/PANewsCN/status/1983127032595898682 Claude ran long-only with a 38% win rate and 25% return, a 2.1 risk-reward ratio, and sub-1 profit expectancy, indicating disciplined but inflexible execution. Source: PANews on X, Oct 28, 2025 https://twitter.com/PANewsCN/status/1983127032595898682 Grok4 briefly led with over 50% gains but reverted to breakeven after frequent flip-flopping and a 20% win rate, underscoring directional inconsistency. Source: PANews on X, Oct 28, 2025 https://twitter.com/PANewsCN/status/1983127032595898682 Gemini executed 165 trades in 10 days, incurred over $1,000 in fees, held a 25% win rate, and earned under $100 per trade on average, exemplifying fee drag and overtrading. Source: PANews on X, Oct 28, 2025 https://twitter.com/PANewsCN/status/1983127032595898682 GPT-5 recorded a 20% win rate with sub-1 risk-reward and chronically late entries, producing outcomes similar to Gemini. Source: PANews on X, Oct 28, 2025 https://twitter.com/PANewsCN/status/1983127032595898682 Key trading takeaways highlighted by @PANewsCN: top performers were low frequency, longer hold, high risk-reward with timely entries, while losers were high frequency, short-term, low risk-reward with late entries; equal information access did not determine profitability; longer reasoning chains correlated with stricter decision-making, with DeepSeek the longest; copying AI trades may not persist across regimes, though AI execution is valuable. Source: PANews on X, Oct 28, 2025 https://twitter.com/PANewsCN/status/1983127032595898682

According to @business, DeepSeek is pushing into Africa, offering cheaper AI technology to entrepreneurs and startups, source: Bloomberg. The report frames the push as making AI more accessible to millions across the region, highlighting a pricing-led expansion strategy, source: Bloomberg. For trading context, market participants in AI and crypto-AI verticals can track the geographic rollout and the low-cost positioning highlighted in the report for signals on competitive dynamics, source: Bloomberg.

According to the source, a public X post on Oct 21, 2025 reports that an AI crypto trading competition showed DeepSeek at $3,650 in unrealized profits and Grok at about $3,000, outperforming ChatGPT and Gemini in the event’s standings. Source: public X post on Oct 21, 2025. The post does not disclose methodology, trade logs, timeframe, realized PnL, or risk metrics, meaning the performance claims cannot be independently verified or audited by traders at this time. Source: public X post on Oct 21, 2025. For trading decisions, treat these figures as unaudited results and avoid extrapolating strategy edge until transparent transaction records or third-party verification are provided. Source: public X post on Oct 21, 2025.