Gemini 4 Carbon Tops the AI Index, DeepSeek Raises ¥105B, Agibot Pitches 'Worken'
The weekly iteration cadence of flagship models, half-price token launches, and multibillion-dollar funding rounds directly reshape cost and capability expectations for any developer building on these APIs. Agibot's shift from shipment counts to Worken hints that evaluation of AI systems is moving from benchmark scores to real-world deployment metrics, a change that will affect how both models and robots are procured and compared.
Google's Gemini 4 Carbon launched just 15 days after its predecessor and immediately topped the Artificial Analysis Intelligence Index with a score of 59, beating Claude Opus 5.5 (58) and GPT-6 Astra (57). Its long-context results are dramatic: 96.2% on GraphWalks versus 71.8% for Astra and 66.8% for Opus, and launch token pricing is at half the usual rate, signaling an aggressive market grab. The rapid iteration cycle shows closed labs compressing releases to weeks, and long-context-plus-reasoning has become the default bar.
On the capital side, DeepSeek's second round came in at ¥105.5B from ~30 investors, with Tencent and CATL leading, and state-affiliated capital at about 27% - a clear "domestic compute + semiconductor supply chain" alliance. OpenAI is close to a $30B round at a $1.25T pre-money valuation, a near 8x jump from its $157B valuation two years ago. Meanwhile, 01.AI received hundreds of millions from Hangzhou state capital in exchange for deploying industrial AI in local scenarios, and Agibot used its Innovation Day to propose a new metric: Worken (effective intelligent work hours), replacing shipment counts as the measure of value. Agibot reports 99.99% success rate and 22-second cycle times on 3C lines, with nearly 500 robots serving 10,000+ visitors daily at Chimelong.
The 15-day gap between Gemini Argon and Carbon shows that closed-source labs now iterate on weekly cycles, raising the cost of staying current and pushing open-source models to match pace or fall behind on long-context features.
DeepSeek's round deliberately loads up ¥105B before an IPO, but the mix of state-affiliated investors and semiconductor partners signals that compute access, not model quality alone, is what the market is really funding.
01.AI's 'scenarios for money' deal with Hangzhou is a concrete example of the industry moving from parameter competition to deployment revenue; local governments are becoming the new enterprise customers for AI.
Agibot's Worken metric reframes robot value from production volume to time-in-production, a shift that aligns incentives for purchasers but also makes quality claims easier to audit and harder to inflate.
The contrast between GPT-6 Astra's 18+ second single inference and Agibot's real-time deployment highlights the gap between cloud intelligence and embodied systems, a bottleneck that funding and new benchmarks are just beginning to address.