Token | Trading Volume 【BTC】2.77 billion 【XPL】39.86 million 【LGNS】7.04 million 【BEAT】15.38 million 【TIA】3.75 million 【JU】182.32 million 【ZAMA】29.48 million 【UNI】13.73 million 【CORE】808,200.56 【SPACE】11.78 million

According to Bijie.com, Google's research team released Gemini-SQL2, a text-to-SQL technology, on June 12th. Built on the Gemini 3.1 Pro model, the system achieved an execution accuracy of 80.04% in the BIRD single-model benchmark for mainstream database queries, setting a new performance record. Earlier versions of Gemini-SQL, running Gemini 2.5 Pro and employing multi-task supervised fine-tuning, previously led the benchmark with 77.14% accuracy; the new version improves upon this by 2.9 percentage points. BIRD, as an industry benchmark for evaluating the capabilities of text-to-SQL systems, includes over 12,700 questions and SQL statement pairs. The test covers real-world database environments such as multi-table joins, dirty data handling, and complex business context reasoning, using the correct execution of the generated SQL statements in the database as the performance metric. Google plans to gradually integrate these capabilities into existing data services, including platforms such as BigQuery Studio, AlloyDB AI, and Cloud SQL Studio, to assist non-technical personnel in directly querying enterprise databases using natural language.

According to CoinWorld, Ramp has released its private benchmark, Ramp SWE-Bench, for cutting-edge AI coded agents. This benchmark includes 80 backend development tasks derived from Ramp's real-world production environment, aiming to address data leakage and metric saturation issues caused by model pre-training in public evaluation datasets. Based on the published cross-model evaluation results of 14 models, Anthropic's latest Claude Fable 5 topped the list with a resolution rate of 87.5%, followed by Claude Opus 4.7 and GPT-5.5 tied for second place with a resolution rate of 83.75%. The test data also reveals the trade-offs between price and performance for different models. The domestic model Kimi K2.6 and GLM 5.1 have similar resolution rates, at 72.5% and 71.25% respectively, but Kimi K2.6 has an average cost of $0.69, approximately 34% cheaper than GLM 5.1.

According to Onchain Lens monitoring, a whale (a large investor) has taken a long position of 54,986 SPCX with 5x leverage, with a position value of $9.09 million. This whale also holds a short position of 320.74 BTC with 25x leverage, with a position value of $20.4 million, currently showing a floating profit of over $1.18 million.

According to CoinWorld, Mistral AI is in talks with investors for a new round of funding, aiming to raise approximately €3 billion (about $3.5 billion) at a valuation expected to reach €20 billion, potentially even higher depending on investor demand. This would be a significant doubling of its €11.7 billion valuation during its September funding round last year. Negotiations are still in the early stages, and specific terms are subject to change. Semiconductor equipment giant ASML is Mistral AI's largest shareholder, having invested €1.3 billion last year for an 11% stake. Founded in 2023 by former researchers from Google DeepMind and Meta, Mistral AI primarily serves European governments and businesses, managing local data centers in France and Sweden. It recently signed cooperation agreements with Airbus and BMW. To expand its business, Mistral AI is in talks with European banks and other institutions, planning to offer cybersecurity models to replace Anthropic's Mythos model.