StepFun Step 3.5 Flash
Hacker ApprovedRanked 25 of 72 language models by recommenders
Fast, cheap open-weight model that people find capable for coding, tool calling and personal assistant tasks. Common caveat: poor token efficiency and occasional hallucinations.
Fast, cheap open-weight model that people find capable for coding, tool calling and personal assistant tasks. Common caveat: poor token efficiency and occasional hallucinations.
53 people gave an opinion
- recommend it
- 40
- criticize it
- 13
- approval share
- 75%
Recommended in 9 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: recommended
- March 2026: recommended
- April 2026: recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 40 people recommend itNeeds 10 or more, met.
- 3.1 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 9 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Low cost per token
“StepFun is a good enough model for tool calling that so far has been incredibly cheap that I'm happy with it.”
Comment from 16 Apr 2026 in “Is anyone actually using OpenClaw?” “Right now I'm really jazzed about how cheap StepFun 3.5 Flash for these basic personal assistant tasks”
Comment from 11 Apr 2026 in “Open Source LLM Comparison – Is Opus Cooked?” “Which isn't super suprising as StepFun is ~about 5% the price of Sonnet.”
Comment from 1 Apr 2026 in “StepFun 3.5 Flash is #1 cost-effective model for OpenClaw tasks (300 battles)” - Comment from 1 Apr 2026 in “StepFun 3.5 Flash is #1 cost-effective model for OpenClaw tasks (300 battles)”
- Comment from 17 Feb 2026 in “Claude Sonnet 4.6 System Card”
Fast generation speed
“being able to boost Step 3.7 Flash up to 40tk/s on my Spark-alike with proper triple head MTP was the thing that made it superior to Qwen 3.6 27B in wall clock time”
Comment from 26 Jun 2026 in “Previewing GPT‑5.6 Sol: a next-generation model” “The official Q4_K_S gguf is quite good and has very good 35 tps generation on a M1 mac studio.”
Comment from 29 May 2026 in “Step 3.7 Flash” “Strong performance, with speed as the highlight—blazing fast, peaking at 350 tokens/s.”
Comment from 2 Feb 2026 in “A new local LLM king: Step-3.5-Flash-int4” - Comment from 19 Feb 2026 in “Step 3.5 Flash: Fast Enough to Think. Reliable Enough to Act”
Runs on personal hardware
“What’s wild about Step-3.5-Flash isn’t just the quality, it’s how close it is to fitting on personal hardware.”
Comment from 11 Feb 2026 in “New open source model achieves same score as GPT 5.2 High on AIME2026 I” “The official Q4_K_S gguf is quite good and has very good 35 tps generation on a M1 mac studio.”
Comment from 29 May 2026 in “Step 3.7 Flash” “surprisingly happy to chat about Chinese banned topics for example when run locally”
Comment from 21 Jul 2026 in “Who's Afraid of Chinese Models?” - Comment from 2 Jun 2026 in “Microsoft's MAI-Code-1-Flash Scores 51% SWE-Bench Pro with Just 5B Active Params”
Strong coding and instruction following
“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.”
Comment from 15 Jun 2026 in “Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?” “conformance to instructions and output quality was better than most of the leading models I've tested”
Comment from 19 Feb 2026 in “Step 3.5 Flash: Fast Enough to Think. Reliable Enough to Act” “For one-shot coding tasks, I found Step 3.5 Flash to be stronger even than Qwen 3.5 397B.”
Comment from 2 Apr 2026 in “StepFun 3.5 Flash is #1 cost-effective model for OpenClaw tasks (300 battles)” - Comment from 24 Jun 2026 in “GLM-5.2 is the step change for open agents”
What people warn about
Each mark is one comment
Verbose output uses many tokens
“it's an amazingly capable model all things considered, but it's token efficiency is so bad”
Comment from 21 May 2026 in “Qwen3.7-Max: The Agent Frontier” “token efficiency is so bad that it gets out performed by most others wall-clock time”
Comment from 20 May 2026 in “Qwen3.7-Max: The Agent Frontier” “The output was fast, but because it wrote so I had to wait longer.”
Comment from 19 Feb 2026 in “Step 3.5 Flash: Fast Enough to Think. Reliable Enough to Act” - Comment from 19 Feb 2026 in “Step 3.5 Flash: Fast Enough to Think. Reliable Enough to Act”
Hallucinates and makes mistakes
“it hallucinates during a lot of tasks, and the output quality isn't something to write home about.”
Comment from 3 Aug 2026 in “Is Hermes Agent unusable without a model router?” “it makes some grave mistakes (had some Hanzi in the output once and typos in words)”
Comment from 1 Apr 2026 in “StepFun 3.5 Flash is #1 cost-effective model for OpenClaw tasks (300 battles)” “Hallucinates like crazy. use with caution.”
Comment from 19 Feb 2026 in “Step 3.5 Flash: Fast Enough to Think. Reliable Enough to Act” Poor name and website
“you really can't see why they wouldn't reach mass market with this name”
Comment from 30 May 2026 in “Step 3.7 Flash” “It's hard to take it seriously due to its name and crappy website”
Comment from 29 May 2026 in “Step 3.7 Flash” “Have to use browser translation which distorts the website”
Comment from 30 May 2026 in “Step 3.7 Flash”
Other language models people recommend
The four most recommended other language models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Claude Opus | 3,331 | 2,746 | 55% | 13 | 4 mo156 owners | — |
| OpenAI GPT | 1,205 | 987 | 55% | 13 | 6 mo47 owners | — |
| Claude Sonnet | 854 | 586 | 59% | 13 | 6 mo46 owners | — |
| DeepSeek V4 Flash | 828 | 223 | 79% | 8 | 1 mo24 owners | Approved |
| StepFun Step 3.5 FlashThis page | 40 | 13 | 75% | 9 | — | Approved |
What people said
40 of the 83 comments that mention it, newest first. Each links to the original comment.
“Step 3.5 was the first model that made me sit up and realise that LLMs weren't a fad”
“Unless you meant step-3.7-flash , the input cache hits are $0.05 per mil”
“I'll be sticking with StepFun 3.7-Flash for the foreseeable future”
“it hallucinates during a lot of tasks, and the output quality isn't something to write home about.”
“surprisingly happy to chat about Chinese banned topics for example when run locally”
“being able to boost Step 3.7 Flash up to 40tk/s on my Spark-alike with proper triple head MTP was the thing that made it superior to Qwen 3.6 27B in wall clock time”
“Step 3.7 Flash locally happily tells me about the Tiananmen Square massacre”
“models are good enough for so many coding tasks even at the lower end”
“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.”
“Step 3.7 Flash on my Asus GB10 based mini pc is incredibly close to that today”
“you really can't see why they wouldn't reach mass market with this name”
“Have to use browser translation which distorts the website”
“Stepfun 3.5 was my daily driver, and I've been blown away at how good it is”
“That was very impressive. Very cool to see more vendors pushing towards models understanding images”
“It's hard to take it seriously due to its name and crappy website”
“The official Q4_K_S gguf is quite good and has very good 35 tps generation on a M1 mac studio.”
“it's an amazingly capable model all things considered, but it's token efficiency is so bad”
“token efficiency is so bad that it gets out performed by most others wall-clock time”
“StepFun is a good enough model for tool calling that so far has been incredibly cheap that I'm happy with it.”
“Right now I'm really jazzed about how cheap StepFun 3.5 Flash for these basic personal assistant tasks”
“For one-shot coding tasks, I found Step 3.5 Flash to be stronger even than Qwen 3.5 397B.”
“$0 is pretty hard to beat in terms of cost effectiveness.”
“it's actually pretty good at openclaw type of tasks for non technical users”
“it makes some grave mistakes (had some Hanzi in the output once and typos in words)”
“I'm not aware of other AI labs that released base checkpoint for models in this size class.”
“Which isn't super suprising as StepFun is ~about 5% the price of Sonnet.”
“if I had to pay for it myself ofc I'd choose stepfun every time”
“StepFun 3.5 Flash is better compared to google's gemini 3 flash”
“I would prefer StepFun-3.5-Flash to Sonnet 4 strongly and to 4.5 often enough”
“I have been using a quorum composed of step-3.5-flash, Kimi k2.5 and glm-5 and I have found it outperforms opus-4.5 at a fraction of the cost”
“conformance to instructions and output quality was better than most of the leading models I've tested”
“it's very fast. It spent some time thinking but does fewer calls”
“The model itself is good but even if you get twice as many tokens as with another model”
“The output was fast, but because it wrote so I had to wait longer.”
“Hallucinates like crazy. use with caution.”
“Beats Kimi K2.5 and GLM 4.7 on more benchmarks than it loses to them.”
“Claude 4.6 is actually worse than the free Stepfun 3.5 version”
“the very best model you can run right now which is Step 3.5 Flash”
“What’s wild about Step-3.5-Flash isn’t just the quality, it’s how close it is to fitting on personal hardware.”
“Strong performance, with speed as the highlight—blazing fast, peaking at 350 tokens/s.”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.