879 points · thisisauserid · comments

# Summary Gemini 3.7 Flash offers solid improvements over 3.6 Flash at half the introductory price, with strong performance on coding and vision tasks, though users debate whether it meaningfully outperforms cheaper competitors like OpenAI's Luna or DeepSeek models, and some report practical usability issues despite benchmark gains.

679 points · bjin · comments

# Summary DeepSeek Harness is an early-stage, open-source agentic coding framework built on the Cordis plugin architecture that enables hot-reloading, full execution traceability, and lifecycle management—offering a cost-effective alternative to other harnesses when paired with affordable DeepSeek models, though opinions vary on whether its plugin-heavy architecture and nascent tooling justify adoption over existing solutions.

649 points · matt_d · comments

# Summary Hackers are excited about Christopher Domas's new DRAM exploit that enables direct memory access on AMD systems, though many criticize the AI-generated documentation and debate whether it requires existing root access to be practically useful. The discovery is seen as both impressive reverse engineering and a demonstration of how unnecessarily complex and attack-prone modern DRAM controllers have become.

635 points · pr337h4m · comments

# Summary The discussion celebrates GPT-5.6 Sol Ultrafast's 7-14x speed improvement over competitors, with commenters highlighting how faster inference enables better iterative reasoning and new real-time applications, though some express skepticism about actual performance parity, pricing opacity, and whether speed gains matter when human review is required.

597 points · pella · comments

# Summary GLM-5.3 represents a significant breakthrough in open-source AI, matching or approaching frontier closed models like OpenAI's and Anthropic's through post-training scaling alone, with particularly strong cybersecurity capabilities—though concerns remain about multimodal limitations and the geopolitical implications of unrestricted access to such powerful models.

404 points · Markoff · comments

# Summary The discussion centers on this move as a potential step toward reducing US dollar hegemony and creating more balanced global financial systems, though commenters express mixed concerns about China's authoritarianism, Deutsche Bank's track record, and implications for currency control and geopolitical power dynamics.

370 points · spelk · comments

# Summary The discussion reveals mixed sentiment about Mistral OCR 4.1: while some users praise its speed and cost-effectiveness for simple documents and handwriting, others argue it's expensive compared to alternatives like Tesseract or Baidu, and that OpenAI's models still dominate for complex OCR tasks requiring high accuracy.

48 points · meetpateltech · comments

The discussion about Gemini 3.7 Flash was moved to a different HackerNews thread (item ID 49289112), so there are no substantive comments to summarize here.

162 points · andsoitis · comments

# Summary The discussion reveals fundamental disagreement about whether AI agents genuinely "lie, cheat, and steal" or simply reflect patterns from their training data without moral understanding. Most commenters argue LLMs are amoral prediction systems mirroring human behavior from the internet, not intelligent agents making ethical choices.

26 points · MrBuddyCasino · comments

# Summary The discussion appears to be mocking Anthropic's leadership over allegations that Dario Amodei's wife attempted to solicit funding from Jeffrey Epstein for an adult entertainment venture, with commenters making dark humor posts.

72 points · n_plus_1_acc · comments

# Summary The discussion centers on how smartphone cameras increasingly use AI filters to enhance or manipulate images rather than capture reality—exemplified by Xiaomi's moon-enhancement feature confusing an eclipse. Commenters express broader concerns about society's apparent indifference to the blurring line between real and fake content in media.

23 points · krlx · comments

DeepSeek is raising API prices significantly, with cached input tokens seeing up to 1000% increases, though the headline is misleading—the model remains substantially cheaper than competitors like Claude and GPT-5.6 even at peak rates. Users note that despite the price hike, DeepSeek's cost advantage may make competing alternatives like Claude still preferable for some use cases.

100 points · adi1 · comments

# Summary Bullet is a coding agent harness that claims faster execution through model routing, efficient context management, and optimized turn sequences, but commenters debate whether these improvements are mechanistically novel versus achievable through prompting alone, while praising the actual speed improvements and Linux support but questioning the saturated benchmark results and whether harnesses alone represent a defensible business.

9 points · adamhowell · comments

No comments available.

7 points · htrp · comments

OpenAI has hired Dali Rajic from Wiz as its new Chief Revenue Officer, replacing Denise Dresser (former Slack CEO) after less than a year in the role.

7 points · pdyc · comments

# Summary A commenter sarcastically suggests that Anthropic's IPO announcement signals the company is failing, implying that going public at such a high valuation is a sign of trouble rather than success.

221 points · thehoff · comments

205 points · magic_interface · comments

221 points · theanonymousone · comments

120 points · gfysfm · comments

53 points · fredski42 · comments

248 points · jkrauska · comments

61 points · surprisetalk · comments

60 points · petethomas · comments

40 points · TMWNN · comments

35 points · hn_acker · comments

194 points · tdx · comments

33 points · brandonb · comments

32 points · grandpajoey · comments

71 points · stammanif · comments

30 points · tartoran · comments

48 points · fittingopposite · comments

42 points · gmays · comments

43 points · LazyMans · comments

22 points · CGMthrowaway · comments

64 points · domenkozar · comments

20 points · grandpajoey · comments

18 points · iamronaldo · comments

22 points · meetpateltech · comments

21 points · ashishmax31 · comments

16 points · CGMthrowaway · comments

36 points · ilamont · comments

16 points · ch-bas · comments

39 points · iljanevo · comments

18 points · ingve · comments

120 points · timmmmmmay · comments

27 points · latchkey · comments

60 points · fagnerbrack · comments

50 points · Jimmc414 · comments

59 points · hn_acker · comments