Ask HN: What LLM models are you using and why?

5/17/2026 at 5:06:00 PM

1. gpt-5.5-medium for most demanding coding tasks.

2. gpt-5.3-codex-medium for genrally most of the other coding tasks.

3. deepseek-v4-flash for heavy agentic research/loops (non-coding related).

4. mimo-v2.5-pro for crunching/summarizing large texts.

5. gemini-3.1-flash-lite for image understanding.

6. opus-4.7 very occasionally when gpt-5.5 fails, or vice-versa, and sonnet-4.6 when codex-5.3 fails.

7. deepseek-v4-pro when I need to do a long agentic session, and want higher quality, for cheap (non-coding).

8. perplexity/pplx-embed-v1-0.6b for embeddings, via openrouter.

by freakynit

5/17/2026 at 6:40:00 PM

I use Opus 4.7 for personal stuff (basically for everything), but have been considering gpt-5.5 given all I hear about it.

At work I use 4.6 because we don't have 4.7 yet...zzz...

I also do a LOT of personal/portfolio work with self-hosted models.

Ministral-3-14B-Reasoning for validating concepts, MVPs, etc and some prod systems (punches above its weight class). Qwen3.6-35B-A3B for self-hosted coding (custom harness). GPT-OSS-120B for self-hosted coding or more reasoning-intensive agentic flows. Qwen3.5-122B-A10B currently in evals for agentic coding.

by zambelli

5/17/2026 at 1:44:10 PM

GPT 5.5 main opus 4.7 frontend and when i need something different kimi 2.6 and GLM 5.1 when i don't have to pay on the task (using the opencode go subscription)

by yossuf2000

5/17/2026 at 8:20:08 PM

By default, I use Sonnet 4.6, and if Sonnet 4.6 fails, I use Opus 4.7.

by teppeik

5/17/2026 at 5:17:42 AM

GPT-5.5 + Opus-4.7 here. Codex for pure coding task with clear goal, claude code for the rest. Also combined with opencode to experiment new models.

fyi: I didn't have much lock on Deepseek v4 pro, with opencode + openrouter it's incredibly slow. How did op did it?

by david_d8912

5/17/2026 at 10:40:31 AM

GPT-5.5 daily.Opus for hard stuff.Deep seek for long context+ cheap iteration.Everthing else is routing and tool now.

by late_night_fix

5/17/2026 at 11:13:07 AM

GPT-5.5 for daily ideas and brainstorming. It has become my daily go to.

by VishnuTech

5/17/2026 at 1:17:29 PM

Opus 4.6. Does the job. Not much of an experimenter.

by dnnddidiej

5/17/2026 at 5:05:19 PM

Opus 4.7 in Vscode via Claude Code.

by enceladus06

5/17/2026 at 7:24:39 AM

for idea using gpt3.5,claude for coding and also impressed by deepseek it large context window is really useful for long projects.

by farwaabbas

5/18/2026 at 12:57:55 AM

qwencodernext. ask me about what it does and doesnt do.

by cyanydeez

5/17/2026 at 7:32:26 AM

[flagged]

by Jbunga

5/17/2026 at 5:47:24 AM

[flagged]

by jaspanglia

5/17/2026 at 8:23:08 PM

[dead]

by unchainedsky90

5/17/2026 at 8:18:04 AM

[dead]

by heji631