Libraries and frameworksOther softwareMedia and creativeAI tools and appsOperating systems and desktopTerminal and shellInfrastructure and DevOpsBrowsers and extensionsEmail and chatEditors and IDEsNetworking and VPNSecurity and passwordsDatabasesAI coding toolsProductivitySelf-hosted and home serverFiles, sync and backupGit and code hostingNotes and knowledgeAI modelsSearchMonitoring and analyticsFinance
AI models
What HN commenters recommend, ranked by distinct recommenders. Under each: what people praise most and the most common caveat, or the strongest quote where there is no summary yet.
One row per model family, counting each person once across its tiers and versions.
Sentiment by version how HN received each release, latest first, across every model →- 1.Latest Claude Opus 5.5: 53% recommend · 12% warn · 203 peopleLoved for Coding ability, Legacy revival · Caveat Safety flags
- 2.Latest GPT Sol 6.1: 49% recommend · 9% warn · 45 peopleLoved for Instruction following, Coding help · Caveat Prose writing
- 3.Latest Gemini Flash 3.8: 56% recommend · 16% warn · 43 peopleLoved for General assistant, Flash value · Caveat Weak agentic coding
- 4.Latest Qwen 3.8: 51% recommend · 8% warn · 225 peopleLoved for Local speed, Coding and agents · Caveat Overthinking
- 5.Latest DeepSeek V4 Flash 4.1: 58% recommend · 4% warn · 113 peopleLoved for Low cost, Good enough · Caveat Insecure code
- 6.Loved for Coding help, Everyday assistant · Caveat Answer quality
- 7.Latest GLM Flash 5.3: 55% recommend · 10% warn · 73 peopleLoved for Good enough, Coding · Caveat Verbal tics
- 8.Latest Kimi 3.1: 41% recommend · 18% warn · 22 peopleLoved for Price for performance, Top open-weight model · Caveat Token usage
- 9.Latest Gemma 4: 47% recommend · 11% warn · 244 peopleLoved for Runs locally, Writing quality · Caveat Compaction failures
- 10.Latest Grok 4.7: 11% recommend · 0% warn · 9 peopleLoved for Coding help, Value · Caveat Political baggage
- 11.Latest Mistral Large 4: 24% recommend · 32% warn · 37 peopleLoved for Low cost, European sovereignty · Caveat Behind frontier
- 12.Loved for Speed, Daily dictation · Caveat Multilingual accuracy
- 13.Loved for Cost and speed, Accuracy · Caveat Specialized models win
- 14.Latest MiniMax 3: 52% recommend · 5% warn · 21 peopleLoved for Cheap token plans, Agent utility roles · Caveat Slow locally
- 15.Loved for Image editing, Home mockups · Caveat Output flaws
- 16.Loved for Runs locally, Speed · Caveat 20B mistakes
- 17.Latest MiMo Pro 2.6: 64% recommend · 27% warn · 11 peopleLoved for Low cost, Coding and specs · Caveat Overthinking
- 18.Latest Muse Spark 1.3: 54% recommend · 8% warn · 13 peopleLoved for Errand automation, Ongoing monitoring · Caveat Privacy
- 19.Loved for Runs locally, Transcription quality · Caveat Outdated
- 20.Latest Cursor Composer 2.5: 66% recommend · 9% warn · 35 peopleLoved for Speed, Low cost · Caveat Complex tasks
- 21.I love how detailed and transparent the data set statistics are on the huggingface pages.
- 22.Loved for Natural voice quality, Audiobooks and articles · Caveat Pronunciation errors
- 23.Latest Llama 4: 22% recommend · 67% warn · 9 peopleI've actually seen better performance with older ones like Meta Llama 4 or Amazon's Nova
- 24.Olmo and to a lesser extent Nemotron are what you should look at
- 25.I ran my custom agentic SQL debugging benchmark against it and I'm impressed.
- 26.Latest Ornith 1.5: 70% recommend · 10% warn · 10 peoplemy new daily driver is Ornith-1.5-35B-A3B-MLX-4bit. I recommend giving that a whirl
- 27.Step 3.7 Flash at IQ4_XS beats Qwen 3.6 27B despite the latter having MTP, on all of my actual coding task evaluations in real codebases.
- 28.Latest Mercury 2.5: 63% recommend · 0% warn · 8 peopleMercury 2.5 is truly groundbreaking, giving it a try is highly recommended.
- 29.When I swapped SAM 2 for SAM 3 the whole system improved
- 30.AlphaFold is great...to come up with a starting model for a chimeric fusion or something.
- 31.granite4 for speed and lower complexity (around the same as gpt20b).
- 32.I use Z-Image Turbo daily -- it has replaced my stock photo subscriptions, for realism and illustrations.
- 33.I find myself reaching for it over proprietary models.
- 34.Switzerland, through EPFL, ETH Zurich, and the Swiss National Supercomputing Centre, has released a complete pipeline with all training data
- 35.showing an incredible level of image/video understanding and generation capability
- 36.the LFM2 models run pretty fast on a phone and are surprisingly capable
- 37.It one-shotted generation of Java bindings for this project
- 38.If you want something different that is designed for levels, check out Marble by World Labs.
- 39.Genie and Veo are definitely closer to her vision than anything World Labs has released publicly
- 40.Hy3 isn't as fast, but so far it seems to stay on track much more reliably than DS4 Flash.
- 41.I used Needle as a router in https://rcarmo.github.io/projects/memento/
- 42.Very nice, multi modal, largest open weight model that supports audio.
- 43.CLIP should give the images that actually look similar.
- 44.Also, Gliner or similar models for different uses.
- 45.I made moonshine the default because it has the best accuracy/latency
- 46.I'm particularly excited to see a "true base" model to do research off of
- 47.Kling has been my go-to for I2V since the 2.0 days.
- 48.Laya is ModernBert-Large (421M) fine-tuned and has a similar to better performance than TypeSafe Jev
- 49.Apple has some impressive privacy technology for Apple Intelligence
- 50.Talkie[1] presents an interesting case for it being at least possible
- 51.it's still worthwhile to review these models and codebases
- 52.the NVIDIA Personaplex demos [2] are very impressive in this regard
- 53.would be a great starting point for a backport to Lisp 1.5
- 54.I recently used WAN to generate a looping clip of clouds moving quickly
- 55.ModernBERT was the clear winner. It learned faster and performed the task better by a significant margin
- 56.there is TabPFN [1] which also has time series capabilities.
- 57.to show it's possible to build an MVP on open weights without years of work and millions of dollars.
- 58.You run OpenCode with Big Pickle on it with decent performance.
- 59.The best world model research I know of today is Dreamer 4
- 60.You can also download the same model and run it locally without data leaving your machine
- 61.some of the biggest open-weight bangers (Krea 2, Ideogram 4, and the kind of slept-on Boogu-Image 0.1)
- 62.I have seen the same impressive performance about 7 months ago here: https://kyutai.org/stt
- 63.We used DINOv2 as the backbone of our RF-DETR model
- 64.the results were actually good with it being opensource and for how cheap the model is to run.
- 65.i am more excited about the ondevice foundation model update that is coming