Pika Unveils Ultra-Fast MiniMax H3 Version: Generation Steps Reduced From 50 to 6, With Enhanced Image Quality

He Yi released a statement addressing the status of AAVE-related collateral funds, clarifying: The 60 million USDT collateralized in AAVE is not fully frozen; hackers only stole approximately $46 million worth of ETH. Once these assets are unfrozen through law enforcement processes, an additional roughly $14 million in assets will be recoverable.

Beating AI News Flash: French AI company Mistral has officially open-sourced the weights of its flagship model Large 4 (codenamed Le Chonk). The model has a total of 1.05 trillion parameters, with approximately 49 billion activated per inference. After opening API previews on October 6, Mistral conducted three additional weeks of reinforcement learning training, and the version released now is an improved iteration. Mistral reported its Terminal-Bench 4.0 programming score rose from 28% to 41%; FrontierCode 2.0 scored 19.7%, ranking first among open-source models. In Artificial Analysis' programming agent index, Large 4 scored 48%, close to GLM-5.3's 49%, while consuming 47% fewer tokens per average question. However, Artificial Analysis only recorded a 25% score for Terminal-Bench 4.0 under its own test configuration, failing to replicate the official 41% result. In other tests, financial table task Finch's score increased from 67.4% to 73.3%; the agent's success rate in resisting prompt injection attacks rose from 93.3% to 97.1%. Its cybersecurity index stands at 51%, still lower than DeepSeek V4.1 Flash's 57%, with more modest overall capability improvements. Artificial Analysis has just upgraded its intelligence index to v5. On the same metric, Large 4's score rose from 25 in the preview version to 26, below GLM-5.3 and DeepSeek V4.1 Flash's 32 points. But inference efficiency improved significantly: output tokens per question dropped by 39%, and hallucination rate fell from 41% to 28%. API prices were also nearly halved. Based on October 29 evaluation pricing and usage estimates, per-question cost fell from $5.54 to $2.06, a 63% reduction. The model weights have been uploaded to Hugging Face, offering FP8 and NVFP4 versions, supporting self-deployment, fine-tuning, and general commercial use. Separate authorization from Mistral is required to operate a commercial model hosting service.

According to TradingBeats monitoring, a whale has concentrated on adding short positions in US stock index futures over the past half hour. The combined notional value of new short positions opened in Nasdaq 100 and S&P 500 contracts totals approximately $9.51 million, with the S&P 500 position switching from long to short. The account currently holds around $7.29 million in XYZ100 short positions at an average price of $28,796.94, plus roughly $2.4 million in S&P 500 short positions at an average price of $7,646.54, bringing the total value of the two short positions to about $9.7 million.

According to TradingBeats' monitoring, a whale has been steadily adding short positions on Intel (INTC) today, lifting its short position count from around 79,900 contracts to 146,000, an 82.7% increase. Since the start of today, the account has added a net total of roughly 66,100 INTC short positions, with a net sell turnover of approximately $4.58 million. During US pre-market trading hours, the whale further net sold around $2.37 million at an average execution price of $68.07, a 31.3% rise. The account currently holds about 146,000 INTC short positions, valued at roughly $10.058 million, with an average entry price of $79.52, generating an unrealized profit of around $1.549 million. Over the same period, INTC contracts trade at $68.91, some 13.3% lower than the position's overall holding cost. The position uses 10x isolated margin leverage, with a liquidation price of approximately $82.46.

Beating AI Express News: OpenRouter has announced an upgrade to its default inference routing algorithm. Previously, the platform primarily selected inference providers based solely on price; now, it will comprehensively consider speed, stability, and actual call costs, no longer blindly choosing the cheapest service provider. The company tested over 24 million requests, with results showing a 16.9% reduction in first-token response time, a 19% cut in full request latency, and a drop in rate-limiting error requests from 7.1% to 3.8%. The tradeoff is a roughly 6% increase in cost per 1,000 requests. The new algorithm is now live, and users do not need to adjust any configurations. This optimization targets the selection of inference providers behind the same model, rather than the model’s own capabilities.

Changxin Technology announced that it plans to repurchase its own shares through centralized bidding transactions, with a repurchase amount ranging from 1 billion to 2 billion yuan, a maximum repurchase price of 72.69 yuan per share, and the funds will be sourced from its own or self-raised capital. Additionally, the company achieved a net profit of 132.04 billion yuan in the first three quarters of 2026, turning around from a net loss in the same period last year.