English text is being prepared.

RTX Spark laptop tested: How does Nvidia's first local AI laptop chip perform?

heise online's c't 3003 tested the Asus ProArt P14, which is equipped with Nvidia's first laptop chip, the N1X (brand name RTX Spark). The machine supports up to 128 GB of unified memory; the test unit had 32 GB of memory plus a 1 TB SSD and cost 3,300 euros. In llama-bench tests with Qwen 3.8-27B, generation speed was about 14.5 tokens/s, slightly faster than Apple M6 and Strix Halo but slower than M5 Pro; with the MoE model Qwen 3.6-35B-A3B, it reached nearly 61 tokens/s. Prompt processing speed was better than Strix Halo and the M5 Pro and M6, but fan noise under full load reached up to 2.7 sone, and the fans also spin up when idle. The author also tried installing Ubuntu; after manually adding the drivers, both models ran 14% and 9% faster under Linux than under Windows.

Aleph Alpha Releases Kolibri-1: A 78B-Parameter Open-Weight Model for German-Language Reasoning

Aleph Alpha has released the open-weight model Kolibri-1, with 78B total parameters. It uses a MoE architecture with 384 experts and 6 routed experts, activating only 3.46B parameters during inference, with a 256K Token context that can be extrapolated to about 1 million Tokens. The model focuses on German and English, and its tokenizer segments German words more token-efficiently; for example, the German Bundestag glossary requires only 1137 Tokens, whereas the compared DeepSeek-V4.1-Flash requires 1910 Tokens. Pretraining used 20 trillion Tokens, with data current as of June 18, 2026. The author's hands-on test found its answers correct and its German-language reasoning process evident, and stated that the Merlin-Arthur protocol makes the model hallucinate less in benchmarks, but its actual performance still needs to be verified in real-world use.

Surface Laptop Ultra review: NVIDIA RTX Spark's local AI performance

The author tested Microsoft's Surface Laptop Ultra equipped with the NVIDIA RTX Spark N1X SoC. The top configuration of this machine features a 20-core Grace Arm CPU, a 6144-core Blackwell GPU, and 128GB of unified memory, with a starting price of $2,599.99. In the UL Procyon AI text generation test, it scored 2703, about 33% higher than the IdeaPad Pro 5 with an Intel Core Ultra X9 388H; MLPerf Flux 2 Klein 4B image generation reached 26.97 images per minute, about 2.7 times that of the latter. The author also used a local Qwen3.8-27B to drive the Hermes agent, which completed extracting benchmark charts from Excel and generating a Word report in about five minutes.

English translation is being prepared.

English translation is being prepared.

Hands-on comparison of three AI office apps: Doubao Work, Qwen Office, and WorkBuddy

Reporters from The Beijing News Shell Finance conducted a series of hands-on task tests on three AI office apps—Doubao Work, Qwen Office, and WorkBuddy—using the Doubao 2.1 Lite0921, Qwen3.87-Max, and Hy4 preview models, respectively. In the long-horizon task from the "2026 AI Office Sector Research Report," Doubao Work produced 10,000 words in 20 minutes 46 seconds, WorkBuddy produced 25,000 words in 21 minutes 18 seconds, and Qwen Office took over 1 hour to produce 50,000 words but consumed about 1,800 tokens, roughly 60 times that of the other two, requiring the reporter to top up 78 yuan to purchase a membership. In the multimodal tests, Doubao Work's poster and video had the highest degree of completion, and the 3D model generated by WorkBuddy was the closest to the original figurine, but when generating a comic it mistook its own icon for a teal pig face.

How Cerebras Keeps Autoresearch Loops From Going Off Track: Lessons From 71 Experiments

Cerebras had an AI agent run 71 experiments across two task types—training optimization and model compression—and found that when the loop scope is tightened, it reliably produces real conclusions, while once the guardrails are loosened, the agent drifts from the goal within hours. In experiment one, codex-autoresearch-harness was used to wrap Codex in a bash loop; GPT-5.4 and Codex-Spark each independently converged on a learning-rate warmdown schedule within 6 hours, but GPT-5.4 had a proposal acceptance rate of 67% versus only 17% for Spark, showing that proposal quality determines total cost more than single-inference speed. In experiment two, REAP plus INT4 quantization compressed a 717 GB model to 92 GB, with dynamic expert swap-in/swap-out taking 0.151 seconds, but after 12 hours of unattended operation the agent changed to asking a different question on its own; the authors corrected this by clearing context, isolating directories, and adding more frequent validation checkpoints.

Cerebras explains model and reasoning-effort selection for GPT-5.6 Sol, Terra, and Luna

Cerebras published a post introducing the three GPT-5.6 models Sol, Terra, and Luna in the Codex subscription; the three are trained and deployed independently and come with reasoning-effort levels, used to trade off among speed, cost, and intelligence. On price, Terra is half of Sol, and Luna is one-fifth of Sol, for both short and long contexts. The article suggests starting most tasks with Luna and upgrading to Terra or Sol when progress stalls; Sol can reach 750 tokens/s on Cerebras, and it also explains how to use the five reasoning levels in Codex, from Light to Ultra.

Stanford's HomeBody connects GPT-6 Astra to the Unitree G1 to tidy up the kitchen

The Stanford team launched the HomeBody project, connecting GPT-6 Astra to the Unitree G1 to enter a kitchen and execute language instructions; it can gather coffee bags, discard designated cardboard boxes, and, when medication is not in view, find the drawer from memory and retrieve the medicine to hand to a person. The system first has Astra guide the G1 to explore an unfamiliar kitchen and save camera images, LiDAR, and joint poses as spatial memory, then reconstructs a digital kitchen in Isaac Sim, after which Astra calls reusable skills such as navigation, grasping, placing, and opening drawers, with the skill modules handling specific motion planning and feeding back results. The team says deploying to a new kitchen requires no additional training data collection or training of specialized motion policies, but perception, planning, and skill execution run on a single RTX 4090 laptop, Astra makes decisions remotely, the initial reconstruction involves preparation time and API costs, and overheating of the finger servos is also one of the current limitations.

Benchmarking Windows 11 Prism Binary Translation: Performance Loss of x86-64 Programs on Arm

Chips and Cheese author Chester Lam used Geekbench 7 to benchmark Windows 11's Prism binary translation, finding that x86-64 programs execute roughly twice as many aarch64 instructions after translation as they would natively, with performance losses of several generations across all cores. Testing was conducted on a Snapdragon X2 Elite Extreme X2E-96-100 laptop and on Neoverse N1 and N2 instances on Azure; even after bearing the translation losses, the Snapdragon's E-Cores and P-Cores still respectively outperformed the Neoverse N1 and N2. The author also examined the translated code in C:\Windows\XtaCache, noting that Prism performs unnecessary register spills and splits 256-bit AVX operations into multiple 128-bit NEON instructions.

Kimi K3 Complete Developer Guide: 2.8 Trillion Parameters, 1M Context, and Together AI Usage

Together AI has released the Kimi K3 developer guide, introducing Moonshot AI's 2.8-trillion-parameter open-weight model, which uses KDA hybrid linear attention, Attention Residuals, and Stable LatentMoE (896 experts with 16 activated), and supports a 1M token context and native visual input. The guide provides usage details for the OpenAI-compatible API, including reasoning_effort (low/high/max, default max), streaming reasoning_content, structured outputs, tool calling, and dynamic tool loading, and notes that sampling parameters are fixed at temperature 1.0 and top_p 0.95. Pricing is $0.30 per 1M tokens for cached input, $3.00 per 1M tokens for uncached input, and $15.00 per 1M tokens for output, with thinking tokens billed as output.

English translation is being prepared.