DeepMind reorganised its leadership after Demis Hassabis stepped from CEO into a chair role, while chief scientist Jeff Dean also left. Gemini 3.5 Pro has missed three release dates since its May I/O reveal, and independent tests say Gemini 3.6 Flash trails multiple major rivals in raw intelligence. DeepMind is shifting day-to-day control to the CTO, losing key talent, and trying to regain momentum without its main scientific leaders while power concentrates closer to Google headquarters.
Writer launched Palmyra X6, a post-training version of Z.ai’s GLM-5.2, alongside upgrades to its agentic harness to reduce deployment token spending. Writer estimates the combined model and harness changes cut costs by up to 50% for basic tasks. It gives customers a more model-agnostic way to handle complex multi-step work with fewer tokens, starting Thursday.
Google introduced Gemini 3.7 Flash, a coding-and-agent model positioned as a more intelligent “workhorse” than its Gemini 3.6 Flash predecessor. It launched three weeks after 3.6 Flash and is priced at $0.75 per 1M input tokens and $3.75 per 1M output tokens. The update improves coding accuracy on benchmarks, web/app generation in fewer prompts, and knowledge-work reasoning, while being rolled into Gemini Spark and shipping with updated safety safeguards.
Google announced Gemini 3.7 Flash, replacing Gemini 3.6 Flash just three weeks after the prior Flash release rather than delivering the expected 3.5 Pro.
Liquid AI released LFM2.5-VL-3B, a 3B-class vision-language model for on-device use that reads screen content and can call tools from text or images. It reports an average score of 69.4 across 28 vision benchmarks. Availability is broadened by shipping the checkpoint in native, GGUF, ONNX, and MLX formats with day-one runtimes, while commercial use of the LFM Open License v1.0 is free only up to $10M USD annual revenue.
Writer Inc. released its Palmyra X6 flagship model and updated its agent platform for larger-scale, multistep workflows for marketing and revenue teams. Palmyra X6 runs at an average of 52% lower agent platform cost and 48% faster speed compared with its prior setup. The updates let customers run unattended long tasks up to eight hours and adds reporting and governance for admins via Playbooks and Skills to track usage, performance, and spend.
DeepSeek released DeepSeek-V4-Flash-0731, a fine-tuned small model that overtook its own DeepSeek-V4-Pro on independent tests. It scored 50 points on Artificial Analysis’ Intelligence Index (max reasoning), up from 44 for DeepSeek-V4-Pro. The update improves agentic performance and lowers per-task costs through the unchanged V4-Flash architecture with new fine-tuning, with weights available under the MIT license and API prices starting at $0.14 per 1 million input tokens.
SpaceXAI released its Grok 4.6 large language model as its current flagship. Grok 4.6 scored 61 on the Artificial Analysis Intelligence Index. The rollout comes about 1 month after Grok 4.5 and includes an updated training pipeline with longer training plus supervised fine-tuning and reinforcement learning, alongside new $2/$6 per million token pricing.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
