author:witcheer

· 35 artifacts, sorted by favorites. · open in search — combine tags, sort, filter by date →

@witcheer 2026-10-03 — I put together a Hermes Agent profile for Nous Portal users who want to keep spend low. what it sets: - main model: De ♥286
@witcheer 2026-08-14 — Qwen3.8-27B running on my RTX 5090, on llama.cpp! first numbers, Q4_K_M via Unsloth's day-0 GGUF, llama-bench: - decod ♥122
@witcheer 2026-10-04 — the promo board on Nous Portal right now: 24 models at 25% to 88% off their list price, and 9 models that are completely ♥118
@witcheer 2026-09-09 — a model 2.5x bigger than my GPU just posted the highest MMLU on my RTX 5090 board. Qwen3.8-Flash-Next. 125B parameters, ♥108
@witcheer 2026-08-28 — I gave grok 4.6 one task in my Hermes Agent: build a feed-watcher as a proper Python package, write the test suite, make ♥102
@witcheer 2026-04-05 — okay this is terrifying lol someone tested 7 frontier models (GPT-5.2, Gemini 3 Flash, Gemini 3 Pro, Claude Haiku 4.5, ♥102
@witcheer 2026-07-12 — Hermes Agent <> Grok 4.5 low giving me very good results so far, I like it very much ♥101
@witcheer 2026-08-03 — sometimes you just need to see what the data says. the most used model on OpenRouter this week is DeepSeek V4 Flash. ♥91
@witcheer 2026-08-24 — Grok 4.6 is half price on Nous Portal this week! it was built with long-running agents in mind, which is exactly the wo ♥81
@witcheer 2026-10-20 — this is a wonderful win for the local AI space! DeepSeek is doing awesome work by open-sourcing their production engine. ♥78
@witcheer 2026-09-22 — a reminder: a lot of the frontier models are at half their list price on Nous Portal right now. Grok 4.7, GPT-5.6 Sol ♥65
@witcheer 2026-09-07 — quick tip, since I just discovered it myself while chasing better QoL in Discord :) I run Hermes Agent from Discord all ♥63
@witcheer 2026-09-13 — the loudest specific complaint in Alex Cheema's thread on why people don't run local models was prefill: a coding agent ♥57
@witcheer 2026-10-14 — Underdog-Saluki-27B is Qwen3.8-27B packed into a 7.9 GB file, out since 8 October. I ran it through my thinking board on ♥55
@witcheer 2026-09-20 — I finished a board I have been running for 10 days: 10 local models with thinking switched on, on one RTX 5090, on four ♥52
@witcheer 2026-10-07 — Gemma 4 31B solved 51 of the 63 problems in this year's AIME and HMMT on my RTX 5090, the most of nine local models. th ♥48
@witcheer 2026-05-24 — the quality battery I use for every model: 1. JSON extraction 2. code generation 3. logic puzzle 4. system prompt adher ♥47
@witcheer 2026-06-17 — GLM-5.2: 743B MoE, MIT licence, 1M context. the biggest open-weights release in months. I did the arithmetic on whethe ♥46
@witcheer 2026-09-10 — I measured the six llama-server flags everyone argues about, on a 16k prompt, on six models: RTX 5090, Qwen3.8-27B, Gem ♥44
@witcheer 2026-09-11 — I gave two reasoning models 4k, 8k and 16k tokens to think on the same 198 GPQA-diamond questions. RTX 5090, llama.cpp ♥42
@witcheer 2026-07-19 — Qwen 3.8 announced open-weight second only to Fable 5 👀 ♥41
@witcheer 2026-07-17 — epoch 1 done: 8 nights, 54,000 steps, one RTX 5090, and Hermes-4.3-36B has its first EAGLE-3 draft head. release bench ♥40
@witcheer 2026-09-24 — Hermes Agent is driving my Microduck! I typed one sentence: "make the duck walk forward for five seconds then sit" Her ♥38
@witcheer 2026-06-22 — benchmarked Qwable-5-27b-coder on my rtx 5090. a real-trace coding SFT of Qwen3.6-27b, trained on Claude Fable-5 trace ♥37
@witcheer 2026-09-12 — I turned every row on the RTX 5090 board into a fit table for 8, 12, 16, 24 and 32 GB cards: measured peak VRAM at a 16k ♥36
@witcheer 2026-08-11 — Meta is back in open weights: Muse Glimmer, a dense 30B VLM, Apache 2.0, released yesterday. I benchmarked it on my RTX ♥36
@witcheer 2026-09-21 — Grok 4.7 is on Nous Portal, 50% off for one week. if your Hermes Agent runs on a Portal account, switching is one line ♥34
@witcheer 2026-10-20 — Muse Glimmer 30B, Meta's new open-weight agent model, runs fully on one RTX 5090 with room to spare, and keeps its speed ♥28
@witcheer 2026-04-21 — update: installed it on my hermes agent, it is working, I still need to tweak it a bit since I did not exactly got what ♥26
@witcheer 2026-04-20 — kimi k2.6 is open-weight and beat GPT-5.4 on SWE-Bench Pro (58.6 vs 57.7) k2.6 sustained 4,000+ tool calls over 12 hour ♥9
@witcheer 2026-04-27 — two model releases shipped on the same day, april 24. it’s worth putting them next to each other. // openai released GP ♥8
@witcheer 2026-07-24 — (2) the aggregator seat: Grok 4.5 is the most named chair, and both people who explained the choice picked it for cost ♥3
@witcheer 2026-10-20 — @mfranz_on would love to but I can only run Qwen 3.8 27B at max ♥2
@witcheer 2026-07-04 — @SlopToSignal I use different profiles, one is Cloud only with GLM5.2 and one is Local only on my own RTX 5090 with Qwen ♥2
@witcheer 2026-09-09 — @_Abides_ grok 4.6 works like a beast under Hermes ♥1