Open-Source AI Efficiency Surges Amid Energy Warnings and Historic Tech IPOs

Kimi-K2.7-Code Open-Source Coding Model Released, Outperforms K2.6 by Up to 31.5% on Benchmarks with 30% Less Reasoning Overhead

Kimi released and open-sourced Kimi-K2.7-Code, their latest coding model. It improves coding and agent performance over K2.6: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, +31.5% on MLS Bench Lite. Reasoning efficiency is enhanced with 30% lower reasoning-token usage compared to K2.6. The model shows better instruction following and higher end-to-end coding task success rates. A 6x High-Speed Mode is planned for future release. The model is available now via Kimi API and Kimi Code.

vLLM v0.23.0 Released: DeepSeek-V4 Matures Across Backends, Model Runner V2 Expands to Llama/Mistral, and Rust Frontend Grows

vLLM v0.23.0 brings 408 commits from 200 contributors and deepens support for recent models. DeepSeek-V4 received massive hardening with sparse MLA decoupling, TRTLLM-gen attention, EPLB mega-MoE, and sliding-window KV cache retention. Model Runner V2 is now default for Llama and Mistral dense models and adds FlashInfer sampling, breakable CUDA graphs, and pipeline-parallel bubble elimination. The Rust frontend gained streaming generate, dynamic LoRA endpoints, /version and /server_info, plus new tool parsers for InternLM2, Phi-4-mini, and Gemma4. Newly supported models include Gemma 4 Unified (encoder-free), MiMo-V2.5, Step-3.7-Flash, Cosmos3 Reasoner, and Cohere Mini Code. The release also deprecates Transformers v4, unifies reasoning/tool-call parsing, and introduces a multi-tier KV cache offloading framework with an object-store secondary tier.

On June 12, 2026, SpaceX completed its initial public offering on Nasdaq, marking a record mega-IPO. The stock soared in its first hours of trading. The debut sent positive ripples through AI, aerospace, and satellite stocks, testing broader tech sentiment at midday.

SpaceX, Anthropic, and OpenAI are reportedly among a group of major private tech firms planning to go public around the same time in summer 2026. The IPO wave comes as the market reawakens and a new acronym, MANGOS—spanning Meta/Microsoft, Anthropic, Nvidia, Google, OpenAI, and SpaceX—is being used to describe the emerging class of megacap companies. The concurrent listings are seen as a stress test for investor demand and valuations.

The United Nations has issued a warning that by 2030, AI systems will consume twice the electricity that France uses in a year at present. This projected surge will drive up costs for digital services and demands more conscious consumption due to its environmental footprint. Even China's aggressive expansion of clean energy may not be enough to guarantee sustainable AI development, according to the UN. The report highlights the strain AI's growing power demands place on global infrastructure.

SpaceX is set to debut on the NASDAQ stock exchange tonight. The initial public offering price is $135 per share. Based on that price, the company's implied market capitalization is $1.77 trillion. This marks a major transition from a private company to a publicly traded entity.

On June 12, 2026, Kioxia Holdings’ total market capitalization reached 44 trillion yen, surpassing Toyota’s approximately 43 trillion yen to claim the top spot among Japanese listed companies for the first time. The rally was driven by expanding sales of NAND flash memory, fueled by massive AI data center investments from U.S. technology giants. SoftBank Group had briefly overtaken Toyota on June 1, lifted by valuation gains from its stake in OpenAI and its Arm Holdings unit. Kioxia’s ascent highlights the growing economic weight of memory semiconductors in the AI infrastructure boom.

Zyphra has released Zamba2-VL, a family of open vision-language models in three sizes: 1.2B, 2.7B, and 7B parameters. Each model uses a hybrid Mamba2 state-space model combined with a small number of shared transformer blocks, replacing dense attention to achieve near-linear inference scaling. The models pair a Qwen2.5-VL vision encoder with this backbone, supporting single- and multi-image understanding and grounding. On 14 benchmarks, Zamba2-VL shows strong visual counting and document understanding (e.g., 90.9 DocVQA for the 2.7B model) but lags larger baselines on knowledge-heavy reasoning like MMMU and MathVista. Its main advantage is an order-of-magnitude lower time-to-first-token compared to comparable Transformer VLMs, particularly beneficial for long multimodal inputs and on-device deployment. Weights are released under Apache 2.0 license on HuggingFace with inference code available.