Fable 5 absolutely crushed the HTML5 physics contest, but... cost 6x more than Opus 4.8 and 39× more than GLM 5.2 in that test. Test was done on atomic[.]chat, a desktop app that runs LLMs locally. The test asked 4 models to generate self-contained canvas demos with believable motion and collisions. The scenes were not simple animations because every crash needed gravity, force, timing, and contact handling. Outputs: - Fable 5: 62,158 tokens, $3.12 - GPT 5.5: 37,753 tokens, $1.14 - Opus 4.8: 22,280 tokens, $0.56 - GLM 5.2: 36,246 tokens, $0.08show more

New Claude Sonnet 5 performs at GPT 5.5 level... 6x cheaper! We gave 4 models the same prompt: build three self-contained HTML5 canvas scenes with real physics crash demos Prompts: - A car crashes into a brick wall - A wrecking ball destroys a house - A catapult throws a rock at a castle wall Outputs: Sonnet 5: 15,047 tokens, $0.15 Opus 4.8: 23,063 tokens, $0.58 Sonnet 4.6: 25,824 tokens, $0.39 GPT 5.5: 31,152 tokens, $0.94 Sonnet 5 did as well as Opus 4.8 and GPT 5.5 on all three tests. In the wrecking ball test, it beat Opus 4.8. The cable moves smoothly and every hit connects. In the catapult test, it beat GPT 5.5. The rock always lands inside the wall. Sonnet 5 still needs better detail and graphics. But it used fewer tokens than every other modelshow more

Grok 4.5 reached GPT-5.6 Sol level across 3 browser... physics demos for free. Prompts were for robot combat, hydraulic pressing, and a canyon-jumping truck. Test was done on atomic chat (atomic.chat), a desktop app that runs LLMs locally. Each answer had to package rendering and physics inside self-contained HTML5 canvas code. These tasks force generated JavaScript to draw frames, apply gravity, detect collisions, and manage controls. Outputs: GPT-5.6 Sol: 12.9K tokens, $0.51 (~7 min) Grok 4.5: 10.8K tokens, $0 (~5 min) Muse Spark 1.1: 26.8K tokens, $0.12 (~7.5 min) GLM 5.2: 10.9K tokens, $0.02 (~12 min) Physics demos expose a weakness that ordinary coding benchmarks often hide: a model can spend more tokens describing a richer scene while still failing the hard part, which is maintaining force, momentum, collision timing, and believable object behavior across frames. (Note, Grok 4.5 is currently free only through Grok Build and Cursor)show more

Grok 4.5 performed GPT Sol level for free! We... gave 4 models the same prompt: build three self-contained HTML5 canvas scenes with real physics demos Prompts: -robot deathmatch, Tombstone vs Minotaur -a hydraulic press flattening stuff on a conveyor -a semi truck jumping a canyon Outputs: GPT-5.6 Sol: 12.9K tokens, $0.51 (~7 min) Grok 4.5: 10.8K tokens, $0 (~5 min) Muse Spark 1.1: 26.8K tokens, $0.12 (~7.5 min) GLM 5.2: 10.9K tokens, $0.02 (~12 min) Grok 4.5 handled all three scenes genuinely well and got surprisingly close to GPT-5.6 this round. On top of that, it ran on the free tier. GPT-5.6 Sol, the frontier model, put out solid but not standout work. GLM 5.2 rendered all three scenes for pennies, but it came out the roughest of the four. Meta's new Muse Spark burned the most tokens yet still stayed cheap, delivering an average result.show more

1-bit Kimi K3 performs at Opus 5 level on... 3D physics! We ran our Atomic Chat quant of Kimi K3 locally on 4x B200 against three cloud models and gave them all the same task, to build a giant anvil drop test as a single HTML file with real physics Outputs: K3 1bit (local): 15.8K tokens, $0 API cost Kimi K3 (API): 15.3K tokens, $0.30 API cost Opus 5: 22.8K tokens, $0.77 API cost GPT 5.6: 14.5K tokens, $0.72 API cost All four got the physics right. But only Kimi made a working winch. The drum turns and the chain drags the flat car off the pad. Opus 5 drew the most detail, road markings and sparks on the hit. And you can run a model at this level on your own box now. That still feels insane to usshow more

New Hunyuan Hy3 hits Gemini 3.5 quality on physics... for 35x cheaper! We gave 4 models the same prompt: build three self-contained HTML5 canvas scenes with real physics demos Prompts: - A bowling ball knocking down the pins - An air hockey rally that ends in a goal - A pool break scattering the rack Outputs: Hunyuan Hy3: 29,757 tokens, $0.006 Gemini 3.5: 23,300 tokens, $0.21 GLM-5.2: 25,454 tokens, $0.07 DeepSeek-V4: 50,600 tokens, $0.009 Tencent's Hy3 matched Gemini across all three: clean collisions, the puck bounced true, the pins scattered like a real strike, the rack broke with real momentum, nothing clipped or floated. GLM is genuinely strong on pure coding tasks, but the moment the job steps outside clean code it gives way. DeepSeek was the letdown, it burned the most tokens of anyone (50k, almost 2x Hy3) and still turned in the weakest scenesshow more

Laguna S 2.1 performs at GLM-5.2 level on building... popular games with 6x fewer params! We gave three local models the same task: build three popular arcade games that play themselves. Each game is one self-contained HTML file with a bot that plays it. Prompts: – Geometry Dash – Doodle Jump – Air Hockey Outputs: Laguna S 2.1: 10.3K tokens GLM-5.2: 26.4K tokens Hy3: 10.4K tokens Laguna held its quality against a 753B model. We think Laguna's Geometry Dash looked the best of the three, the cube clears every spike and block. GLM won Air Hockey. Its table looked the most detailed of all. Hy3 was the only model that added shooting to its Doodle Jump. But Laguna is the only model in our benchmark that runs on a MacBook with 128GB!show more

Sakana Fugu surprisingly performed near GLM 5.2 level but... 17× more expensive! We gave the same prompt to 4 models: build a complete live Trader Desk with both frontend and backend components, real-time market data fetched from external APIs for 8 symbols, and a custom dark-theme UI. Outputs: Fugu Ultra — 22,225 t, $0.51 Opus 4.8 — 15,802 t, $0.31 GPT-5.5 — 11,474 t, $0.26 GLM 5.2 — 13,677 t, $0.03 Fugu created the most polished and feature-rich trading desk in the run. GLM 5.2 was very close behind, with a similarly complete multi-panel interface and live data, but at a much lower cost. Opus and GPT also performed well, delivering solid results with a better balance between quality and costshow more

HERMES AGENT NOW RUNS CLAUDE OPUS 5. NEAR FABLE... 5 INTELLIGENCE. HALF THE PRICE. SELF-VERIFIES ITS OWN WORK. AVAILABLE TODAY VIA NOUS PORTAL (20% OFF ALL MODELS). Anthropic shipped Opus 5 on July 24, 2026. same $5/$25 per million tokens as Opus 4.8. but the benchmarks tell a different story. WHAT CHANGED FROM OPUS 4.8: FrontierBench v0.1: Opus 5: 43.3%. Opus 4.8: 18.7%. 2.3x jump on the same test. ARC-AGI-3: Opus 5: 30.2%. 3x better than the next closest model. beat Fable 5 on 8 out of 13 benchmarks. at half the cost ($5/$25 vs $10/$50). same price as Opus 4.8. twice the intelligence. no reason to stay on 4.8. THE SPECS: model ID: claude-opus-5 context: 1M tokens (default and maximum) max output: 128K tokens thinking: on by default effort toggle: low / medium / high per request fast mode: $10/$50, 2.5x faster knowledge cutoff: May 2026 minimum cacheable prompt: 512 tokens (was 1,024) SELF-VERIFICATION (the biggest change): Opus 5 checks its own work automatically. Anthropic says: delete your verification prompts. "include a final verification step" now causes OVER-verification because the model already does it. for Hermes /goal tasks this is a direct upgrade. the judge checks evidence. the model also checks evidence. double layer of verification without extra tokens. EFFORT TOGGLE: low: fast, cheap, routine work. medium: balanced, daily tasks. high: full reasoning, complex problems. set per request. not a global switch. matches Hermes /reasoning command: /reasoning low (routine) /reasoning high (complex) Opus 5 effort toggle + Hermes reasoning control = precise cost management per turn. WHERE OPUS 5 FITS IN HERMES: DAILY DRIVER (replaces Opus 4.8): same price. 2.3x better benchmarks. set as your main model: Desktop app / Dashboard: Models → claude-opus-5 CHIEF OF STAFF: synthesis across multiple agents. reads Kanban, prioritizes, routes tasks. self-verification catches routing errors before they cascade. COMPLEX CODING: SOTA on agentic coding benchmarks. FrontierBench 43.3% = best public model for coding. set as coder profile model. /GOAL TASKS: self-verification + completion contracts = the model proves its work AND double-checks the proof. long-horizon goals finish correctly more often. MoA AGGREGATOR: strongest synthesis model at $5/$25. pair with GPT-5.6 and Grok 4.5 as references. Opus 5 aggregates. best quality at mid-range price. presets: max-quality: reference_models: - provider: openai-codex model: gpt-5.6-sol - provider: xai model: grok-4.5 aggregator: provider: anthropic model: claude-opus-5 COMPUTER USE: near-Fable 5 quality for browser automation. at half the token cost per session. computer_use tasks burn lots of vision tokens. Opus 5 halves that bill vs Fable 5. WHAT TO KEEP OPUS 5 AWAY FROM: cron monitoring: too expensive. use DeepSeek or no_agent mode. sub-agent grunt work: use GPT-5.6 Luna ($1/$6) or DeepSeek. auxiliary tasks: use Gemini Flash. routine web extraction: use a cheap model. Opus 5 is for the turns where quality compounds. planning, synthesis, verification, complex reasoning. budget models handle everything else. NOUS PORTAL: 20% OFF ALL MODELS Nous Portal currently runs a 20% discount on all models including Opus 5. $5/$25 official → $4/$20 through Nous Portal. the cheapest way to run Opus 5 right now. hermes setup --portal select claude-opus-5 as your model. discount applies automatically. Opus 5 replaces Opus 4.8 everywhere. same price. better at everything. no tradeoff. straight upgrade. hermes update /model claude-opus-5show more

Sonnet 5 is here. It's worse than Opus 4.8... on nearly every benchmark... Does that mean it's useless? Absolutely not. Use it with Claude Code Dynamic Workflows! 1. /model set to Sonnet 5 2. /effort set to Ultracode 3. Any complex task will kick off a dynamic workflow This will only become more powerful when Fable is back. You'll use Fable 5 as the superintelligent advisor and Sonnet 5 as the fast and efficient implementer.show more

Flappy Bird w/ DeepSeek V4 Pro vs GLM 5.2... vs Fable 5 I ran a small experiment: same game, same prompt, three different models. Test setup was simple: /𝚍𝚎𝚜𝚒𝚐𝚗 in Command Code. The thing I cared about was not UI. UI is now relatively easy for models to imitate. I cared more about UX: whether the model understands the interaction, the flow, the hierarchy, the little product decisions that make something feel usable. 𝙲𝚘𝚜𝚝𝚜: 𝙳𝚎𝚎𝚙𝚂𝚎𝚎𝚔 𝚟𝟺 𝙿𝚛𝚘: $𝟶.𝟶𝟶𝟶𝟾 𝙶𝙻𝙼 𝟻.𝟸: $𝟶.𝟶𝟺𝟾 𝙵𝚊𝚋𝚕𝚎 𝟻: $𝟶.𝟺𝟸 The surprising result was that Fable 5 did not produce a meaningfully better UX. It was good, but not good enough to justify the pricing delta for this kind of design task. Open-source models are getting very interesting here. They are much cheaper, competitive on UI, and increasingly good enough on UX. The gap is closing fast.show more

BREAKING: Anthropic just dropped Opus 4.8—and it is a... MONSTER We've been testing for about a week Every 📧 and our verdict is they could've just called it Opus 5, it's that good. Here's our vibe check: - Beats GPT-5.5 on Senior Engineer bench. On our toughest benchmark Opus 4.8 scores a 63—a hair higher than GPT-5.5's score of 62, and a full 30 points higher than Opus 4.7. It tackled a ground-up rewrite of a production codebase, and actually built something that works. HOWEVER: Coding performance varied a lot at different reasoning levels. We recommend using it on xhigh for best results. - Incredibly good writer. Opus 4.8 scored a 79.6 on our writing benchmark—measuring models on real-world writing tasks we do all of the time like essay writing, promo email writing, and more. It beats GPT-5.5 by 6 points. It produces well-written prose with fewer "AI-isms". It's also very good at writing in your voice given the right context. HOWEVER: Writing performance also varied with reasoning levels. Medium reasoning had higher incidence of AI-isms—we found best results with high. - Beast at knowledge work. Opus 4.8 is very good at general knowledge work tasks like report creation, research and more. It produced the best PowerPoint one-shot we've ever seen on our deck generation benchmark. - Emotionally intelligent, willing to question the frame. I've also found it to be quite good at talking through psychological or interpersonal issues. It has a high EQ, and it's also good at not glazing and helping to expand your perspective. Its thought process feels extremely rich and dynamic. THE BAD: These days a model is only as good as its harness, and Codex is still a far superior harness to the Claude Desktop app. This has kept me using Codex + GPT-5.5 as my daily driver, but I am flipping back and forth a lot more between Codex and Claude. Anthropic is back baby! Read the rest on Every 📧:show more

I can't believe this is real I have GLM... 5.2 running 100% locally on my Mac Studio. 2 bit quant. The results I'm getting are better than Opus 4.8 It's now powering my Hermes Agent and Codex. 100% free, local, private super intelligence on my desk I also have it in a loop coding for me 24/7 now I thought we were at least a year away from this type of event. It happened today. The model takes up about 250gb of memory. So you can technically run it on a Mac Studio with 256gb, but you probably want the 512gb memory version (please tell me you listened to me 5 months ago when these were sitting on store shelves) With Fable gone, I now have Opus 4.8 level intelligence on my desk for free. This is the future. Local, private, secure, personal super intelligence. If you're still writing off local AI as a fad or engagement bait, you are officially delusionalshow more

Most recent diffusion language model research (that I’ve seen)... seems to be using masking as the noising process. It looks like, however, most closed-source models (Google Gemini Diffusion and possibly Inception Labs’ Mercury) use a different noising process, where instead of masking tokens, they replace them with different tokens (either with a random token or a semantically similar token). I wondered how they were getting such high throughput with the latter noising process, since I believed that optimizing inference with KVCache approximation would be more difficult (for various reasons). I visualized this noising process with tiny-diffusion and compared it to normal unmasking, and was very surprised to see how fast the generation “settles” into a reasonable output, and then only slightly refines afterwards, requiring much fewer steps in total. Unmasking (where tokens are never remasked, the typical implementation) is inherently limited in generation speed by the fact that an increase in tokens decoded per step leads to more errors due to the mismatch between individual and marginal token probability distributions we sample from. The token replacement noising process seems to have a much different set of characteristics. Because we sample each token per step, every token makes “progress” towards the final output each iteration (in addition to *potentially* giving other tokens more information in future steps). Generally, masking has outperformed other noising processes, which is probably why most research focused on it (using smaller models). But the paper referred to in the retweet shows that random replacement as a noising process may scale better as model size increases. Big labs might have noticed these results much earlier (due to having drastically more training resources and being able to test larger models), which may explain the discrepancy in the choice of noising process. I’m gonna test this with larger models, since tiny-diffusion only has 10M parameters.show more

Holy moly: GLM-5.3 got much better in cybersecurity since... our pre-release evaluation with Z.ai. It now matches GPT-5.6-Sol on our cybersecurity benchmark at 0.4x the cost 🤯 - At pass@1: it went from 60.4% to 65.6% CVEs rediscovered, crushing every other open model on one-shot tasks - At pass@3: it did 75% -> 78.1%, matching GPT-5.6-Sol - Its precision remained stable, reporting fewer false positives than DeepSeek models The performance increase comes from a behavioral change: the new version is more persistent. It tends to run longer, and had a ~43% reasoning tokens increase. But the performance upgrade is worth that additional cost. 1/3 🧵show more

Right now, you may not have access to models... like GPT‑5.6 Sol, GPT‑4.6 Terra, GPT‑5.6 Luna, Claude Mythos 5, or Claude Fable 5. But you can run something surprisingly powerful today, locally, and completely free. in the next 10 mins on your 8 GB VRAM gaming laptop. Gemma 4 26B A4B QAT (MoE) delivers strong performance on a standard 8 GB VRAM GPU using Ollama, with no API, no usage limits, and no external dependencies. Out of the box, it reaches around 20 tokens per second without any optimizations. Only one command in your terminal: Ollama run gemma4:26b This means: Full offline capability (privacy by default) Zero recurring cost Competitive performance for many real world tasks Fast enough for interactive use on cheap consumer hardware If you're waiting for cutting edge cloud models, you're missing what is already practical today: a capable, local LLM that runs entirely on your own machine.show more

Fine-tune DeepSeek-OCR on your own language! (100% local) DeepSeek-OCR... is a 3B-parameter vision model that achieves 97% precision while using 10× fewer vision tokens than text-based LLMs. It handles tables, papers, and handwriting without killing your GPU or budget. Why it matters: Most vision models treat documents as massive sequences of tokens, making long-context processing expensive and slow. DeepSeek-OCR uses context optical compression to convert 2D layouts into vision tokens, enabling efficient processing of complex documents. The best part? You can easily fine-tune it for your specific use case on a single GPU. I used Unsloth to run this experiment on Persian text and saw an 88.26% improvement in character error rate. ↳ Base model: 149% character error rate (CER) ↳ Fine-tuned model: 60% CER (57% more accurate) ↳ Training time: 60 steps on a single GPU Persian was just the test case. You can swap in your own dataset for any language, document type, or specific domain you're working with. I've shared the complete guide in the next tweet - all the code, notebooks, and environment setup ready to run with a single click. Everything is 100% open-source!show more

Does LLM really need to be a helpful assistant... all the time? No. If you want to simulate people, “perfectly helpful” could be the wrong objective. Meet OdysSim, a journey toward LLMs beyond assistants, as behavioral foundation models (10B tokens of real human behavior; 23 sim benchmarks, finally in one place. new open models: outperform or on par with GPT-5.5, Gemini 3.1, or Claude Opus 4.7 in many behavior-sim dimensions). Human behavior simulation is becoming essential. Agent evaluation needs realistic users before real users show up. Medical and classroom training need realistic patients and students. Social science needs synthetic participants at scale. But real people are not ideal assistants. Real patients panic or ignore good advice. Real students misunderstand. Real customers are vague, picky, impatient, or simply leave. Human behavior is messy, diverse, and often imperfect. Frontier LLMs are getting better at math, code, and long-horizon tasks. They are NOT getting better at simulating human behavior. If anything, they drift the other way: more assistant-ish, more homogeneous, fewer of the errors and quirks real humans show. This is no accident. The whole pipeline is built for helpfulness and task success, not behavioral realism. And you can't prompt your way out of that. So we rethink the recipe from scratch and release: 🧠 The OdysSim corpus: 21.4M real human interactions (~10B tokens) from 62 sources, every conversation retrofitted with social grounding (who is talking, and why) 📏 SOUL-Index: 23 human-behavior benchmarks unified into one suite across 5 axes 🤖 OSim-8B: open weights; tops more SOUL-Index benchmarks than any frontier model, acts more like a real user than any of them on τ-bench (nearly matching real humans in the reaction dimension), and writes far more human-like text along the way.show more

After a few more hours, I think I've figured... out Opus 5. Opus 5 is trained to be more agentic than anything I've used. All Claude 5 models are like that. So what changes? The way to interact with Opus 5 or contextualize it won't work the same way as with other models. It loves exploring, so it doesn't need much guidance for it. Unique preferences, artifacts, and references compliment it well and enable cleaner and more effective exploration and execution. Now that it can explore more effectively on its own and understand intent better, the best thing to do is to get out of its way (e.g., it doesn't need examples of your preferences; a clear high-level description of it works best). It's truly agentic in that sense. A good first step to provide better context for Opus 5 is to distinguish between what's situational and what needs persistence. Regardless, persistent system prompts and CLAUDE.MD needs to stay lightweight. Remove memories and tool descriptions from these. CLAUDE.MD is also a great place to tap into progressive disclosure by linking command/skills to it. On the situational side, agent skills and auto-memory can leverage progressive disclosure and the improved ability of the model to use its external context/knowledge. Conflicting and unnecessary instructions, which are common at this layer (mainly to ensure reliability), are going to throw off this model easily. That's the biggest change I had to make. Simple, clean, and clear prompts and skills work best. I had to clean a lot of my skills and system prompts. The way I prompt remains the same (usually clear and well-scoped). MCP tool descriptions are also more descriptive and have been deduped from the system prompt. Anthropic released a guide on the new rules for context engineering, which was helpful here. I started to test the recommendations and created a little artifact with the things that worked along the way. This might feel like a lot of work. Believe me, it has been frustrating. But I think we can expect future frontier models to become more agentic and smarter at figuring out the right context/gaps. The best thing to do is to prepare for that now. Boris Cherny mentioned that Opus 5 is their least prompt-injectable model yet. I am not sure if that was something they intentionally trained for or if it emerged based on how it was trained, which is to be extremely agentic in nature and more direct in execution.show more

Qwen 3.8 27B Q4_K_M - 90 tokens/sec on a... single NVIDIA RTX 4090 (24 GB VRAM) with Dflash2! (MTP 60 tps -> 90 tps Dflash2!!!!) Local AI moves so fast (literally!) it’s terrifying. Z lab just dropped DFlash 2 for Qwen 3.8 27b and Muse Glimmer. I patched llama.cpp (PR #27342) and paired it with Unsloth’s Qwen 3.8 27B UD-Q4_K_XL quant. The result? Lossless 90 tokens/s decode. My last post highlighted native MTP hitting 60 t/s at 130,000 context. But DFlash 2 just completely shattered that ceiling. By using parallel block diffusion drafting (predicting whole blocks of tokens in a single pass using dynamic convolutions), DFlash achieves a massive 5.39 token acceptance rate. THE ALPHA TWEAK: `n-max 7` eats too much VRAM for draft states. But if you drop the draft limit to `--spec-draft-n-max 4`, you slash the VRAM overhead and actually increase the throughput. Here is the new 24GB VRAM Physics Matrix (DFlash 2 @ n-max 4): - 30k Context: 1,725 t/s prefill | 87.05 t/s decode | 22.2 GB VRAM - 80k Context: 1,789 t/s prefill | 84.20 t/s decode | 23.3 GB VRAM - 110k Context: 1,767 t/s prefill | 83.35 t/s decode | 23.96 GB VRAM (110k context at 83+ tokens a second sitting exactly on the 24GB hardware limit is absolute wizardry). How to compile the PR today: git clone cd llama.cpp git fetch origin pull/27342/head:pr-27342 git switch pr-27342 cmake -B build -DGGML_CUDA=ON && cmake --build build -j Llama.cpp flags for Dflash (110k Context Ceiling): ./build/bin/llama-server -m Qwen3.8-27B-UD-Q4_K_XL.gguf -md Qwen3.8-27B-DFlash2-Q4_K_M.gguf --spec-type draft-dflash --spec-draft-n-max 4 -c 110000 -ngl 99 --port 8080 -ctv q4_0 -ctk q4_0 The fact that the open source community is shipping block diffusion drafters so quickly that run entirely locally on a gaming GPU is unbelievable. If you own a single RTX 3090 or 4090, it is officially time to upgrade to qwen 3.8 27b with dflash 2 and cancel your API subscriptions and let local silicon eat the cloud. This model beats GPT 5.6 Terra, GLM 5.2 DeepSeek V4 Pro, Muse Spark 1.2 and Claude Opus 4.8 on the artificial analysis agentic index (details in the replies) Hugging Face GGUF links (Base + DFlash2) and the full visual VRAM scaling and Dflash2 vs MTP graphs are also in the replies below. are you sticking to native MTP for the 130k context, or sacrificing 20k context to redline your decode speed? How many tokens/sec are you pushing on your current local rig?show more

Qwen3.8-Flash-Next is starting to feel like the local model... Opus fans have been waiting for. Someone ran the NVFP4 176B-class Flash-Next on 2× DGX Sparks, and the results are wild. Real measured scaling → C1: 44.2 tok/s → C2: 64.6 tok/s → C4: 86.8 tok/s aggregate The per-stream speed drops with concurrency, but total throughput keeps climbing. Long-context behavior was even more impressive: → 5K: needle retrieved → 21K: needle retrieved → 84K: needle retrieved → 167K: needle retrieved → 262K: prefill succeeded, but the window was saturated That 167K retrieval test is the one I care about. Long agent runs are where models usually start losing the plot. Flash-Next didn’t. It also held up surprisingly well on physics-heavy reasoning, artifact generation, research workflows, evidence checking, and long-horizon planning. The personality is interesting too. DeepSeek V4 Flash feels like the dependable workhorse. GLM-5.2 feels like the problem-solving machine. Qwen3.8-Flash-Next feels more insightful. It has that rare ability to understand what you’re actually asking rather than just following the surface pattern. The main weakness I’ve noticed is instruction following. It can occasionally drift between prose turns where DeepSeek and GLM stay tighter. And this is why the 256GB M5 Ultra conversation gets interesting. If Apple can pair that huge unified-memory pool with enough bandwidth, this model class becomes genuinely practical for long-running local agents. We’re talking about frontier-class reasoning on hardware sitting on a desk.show more