LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
Step 3.5 Flash
◆ OPEN
by StepFun
OPEN WEIGHTSHOSTEDAPILONG CONTEXTPAID
RANKED #257 OF 318 ASSISTANTS · OVERALL #396 OF 494 RANKED · VIBE SCORE 5.5 · 35 VOICES ·
COPY BADGE pastes a README snippet with the rank SVG.
Step 3.5 Flash is StepFun's most capable open-source foundation model. Built on a sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token....
HAPPY9.9
6 mentions
SAD5.1
1 mentions
VIBE TRAJECTORY
26 weeks · 28 voices in this window
HONEYMOON FADING
Early praise is cooling off in recent weeks.
HOW TO USE IT
Open weights. You can use a hosted service, or download it and run it yourself, free.
RUN IT ON YOUR OWN MACHINE
Open weights: download Step 3.5 Flash once and it's yours. No subscription, no rate limits, works offline.
196B parameters (11B active MoE), a Q4 file is about 111 GB
- BARE MINIMUM
Mac Studio M4 Max (128 GB)
128 GB UNIFIED MEMORY
~$3999 NEWQ4_K_S quant, ~30 tok/s, 32k context
- SWEET SPOT
Mac Studio M3 Ultra (192 GB)
192 GB UNIFIED MEMORY
~$5500 USEDQ4_K_M or Q5 quant, ~36 tok/s, 128k context
- FULL POWER
Custom Rig (4x RTX 4090 24GB + 2x RTX 3090 24GB)
144 GB VRAM
~$9000 USEDFP8 unquantized base, ~45 tok/s, 256k context
LEARN HOW TO SET IT UP
WILL MY BOX RUN STEP 3.5 FLASH?
The three cards above are researched picks. Search the machine you actually have — we only say yes if it should reply at a usable speed, not just load the weights and crawl.
HARDWARE PICKS & PRICES ARE RESEARCHED FROM THE LIVE WEB AND REFRESHED AUTOMATICALLY. TREAT THEM AS BALLPARK, NOT GOSPEL.
WHAT CAN YOU ACTUALLY DO WITH IT?
Everyday jobs on each of those rigs: how long Step 3.5 Flash takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMMac Studio M4 Max (128 GB) | SWEET SPOTMac Studio M3 Ultra (192 GB) | FULL POWERCustom Rig (4x RTX 4090 24GB + 2x RTX 3090 24GB) |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~17 S★★★★★ | ~14 S★★★★★ | ~11 S★★★★★ |
| Summarize a documenta long report boiled down to the points that matter | ~50 S★★★★★ | ~42 S★★★★★ | ~33 S★★★★★ |
| Build a websitea small landing page, markup and styles together | ~4.4 MIN★★★★★ | ~3.7 MIN★★★★★ | ~3 MIN★★★★★ |
| Build a backendan API with routes, storage and tests | ~14 MIN★★★★★ | ~12 MIN★★★★★ | ~9.3 MIN★★★★★ |
| Build a gamea playable browser game in one file | ~8.3 MIN★★★★★ | ~6.9 MIN★★★★★ | ~5.6 MIN★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
WHAT THE INTERNET SAYS
The internet is split on Step 3.5 Flash. Biggest praise: help with code. Biggest gripe: getting facts right.
31 POSTS · 4 COMMENTS · LEMMY · BLOG · REDDIT · X · GITHUB
PEOPLE LOVE
- Popular for coding help (3 mentions)
PEOPLE COMPLAIN ABOUT MORE
- Sometimes makes things up (3 mentions)
TOPIC BY TOPIC
- Help with codePEOPLE LOVE THIS
3 thumbs up · 1 thumbs down
- Getting facts rightA SORE SPOT
0 thumbs up · 3 thumbs down
LOOKING FOR AN ALTERNATIVE? SEE ALL
Other models the crowd has fully reviewed, starting with text models like this one.
Qwen3.8 27B
Muse Spark 1.3 Contributor
DeepSeek V4 Flash> SOURCES (12)
- X · 2026-08-20 · 37 UPVOTESyou can use step 3.5 flash by stepfun for FREE for 15 days😳 stepfun one of chinas ai tigers is running a limited promo
you can use step 3.5 flash by stepfun for FREE for 15 days😳 stepfun one of chinas ai tigers is running a limited promo
- REDDIT · 2026-02-10 · 34 UPVOTESLast week in Image & Video Generation
Last week in Image & Video Generation
- REDDIT · 2026-02-19 · 33 UPVOTESWhy DeepSeek V4 doesn't need more parameters to win
Why DeepSeek V4 doesn't need more parameters to win
- BLOG · 2026-06-06 · 32 UPVOTESLLM Research Papers: The 2026 List (January to May)
As some of you know, I have the long-running habit of keeping a running list of research papers I want to read, revisit, or cite in future articles and projects. Last year, I shared two organized paper lists, one covering January to June and another one…
- BLOG · 2026-03-22 · 32 UPVOTESA Visual Guide to Attention Variants in Modern LLMs
I had originally planned to write about DeepSeek V4. Since it still hasn’t been released, I used the time to work on something that had been on my list for a while, namely, collecting, organizing, and refining the different LLM architectures I have covered…
- BLOG · 2026-02-25 · 32 UPVOTESA Dream of Spring for Open-Weight LLMs: 10 Architectures from Jan-Feb 2026
If you have struggled a bit to keep up with open-weight model releases this month, this article should catch you up on the main themes. In this article, I will walk you through the ten main releases in chronological order, with a focus on the architecture…
- REDDIT · 2026-06-16 · 29 UPVOTESHalBench: 29 OSS models tested on a custom built Sycophancy and Hallucination Benchmark, Qwen 3.6 and Gemma 4 scoring far above their…
HalBench: 29 OSS models tested on a custom built Sycophancy and Hallucination Benchmark, Qwen 3.6 and Gemma 4 scoring far above their weight! (While Meta keeps proving they forgot how to spend their money...)
- REDDIT · 2026-02-19 · 25 UPVOTESGLM-5 is officially on NVIDIA NIM, and you can now use it to power Claude Code for FREE 🚀
GLM-5 is officially on NVIDIA NIM, and you can now use it to power Claude Code for FREE 🚀
- REDDIT · 2026-04-09 · 24 UPVOTESWhen are we gonna get more 1-Bit models(Medium & Large size)?
When are we gonna get more 1-Bit models(Medium & Large size)?
- REDDIT · 2026-02-12 · 16 UPVOTESStep-3.5-flash Unlosth dynamic ggufs?
Step-3.5-flash Unlosth dynamic ggufs?
- REDDIT · 2026-05-04 · 13 UPVOTESWhy is Step-3.5-Flash (196B-A11B) much cheaper to run than Qwen3.6-35B-A3B?
Why is Step-3.5-Flash (196B-A11B) much cheaper to run than Qwen3.6-35B-A3B?
- REDDIT · 2026-08-08 · 5 UPVOTESIs deepseek v4 flash the best model as subagent?
Is deepseek v4 flash the best model as subagent?