Step-3.5-Flash: benchmark scores, price and ranking
Step-3.5-Flash from stepfun-ai ranks #64 of 743 LLMs on the LLMs Tiger aggregate benchmark score (52.5), based on 2 benchmark results captured up to 2026-10-13.
- Organisation
- stepfun-ai
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2026-01-29 (source: openrouter)
- Cheapest API price
- $0.100 in / $0.300 out per million tokens (via openrouter, observed 2026-10-13T20:47Z)
- Aggregate rank
- #64 of 743 (score 52.5)
Across the 2 benchmarks where at least ten models are scored, Step-3.5-Flash is in the top 10% on 0 and in the top half on 2.
Its cheapest listed input price, $0.100 per million tokens, is lower than that of 79% of the 427 models with a listed price.
Step-3.5-Flash benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| MMLU-Pro | 84.4% | #16 of 136 | HuggingFace MMLU-Pro leaderboard | 2026-10-13 |
| GPQA Diamond | 83.5% | #90 of 275 | HuggingFace GPQA Diamond leaderboard | 2026-10-13 |
About Step-3.5-Flash
Step-3.5-Flash is an open-weight model that can run on your own hardware from stepfun-ai. It was released on 2026-01-29, according to the date OpenRouter first listed it (approximate). The cheapest API listing we track charges $0.100 per million input tokens and $0.300 per million output tokens, through openrouter. We hold 2 benchmark results for it across 1 area, and none is in the top tenth of its board.
Where Step-3.5-Flash stands
On our aggregate score, Step-3.5-Flash is #64 of 743 ranked models, in the top quarter. Its score is 52.5; the scale is centred on 50, so it sits 2.5 points above the typical model. That beats 91% of the ranked field.
Step-3.5-Flash, benchmark by benchmark
MMLU-Pro puts Step-3.5-Flash at 84.4%, #16 of 136. That beats 88% of the models tested, but 3.7 points short of Darwin-180B-RSI (88.1%). The results either side are glm-4.5 (84.6%) and DeepSeek-R1 (84.0%). It covers 57-subject knowledge + reasoning. The board median is 55.3%, so Step-3.5-Flash is 29.1 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.
On GPQA Diamond, Step-3.5-Flash reaches 83.5%. It ranks #90 of 275, ahead of 67% of the field, and 12.3 points behind GPT-6 Astra (max) at 95.8%. The results either side are Muse-Glimmer-30B (83.5%) and qwen3.5-35b-a3b (83.5%). It covers expert-level science questions. The board median is 71.7%, so Step-3.5-Flash is 11.8 points above it. Source: HuggingFace GPQA Diamond leaderboard, captured 2026-10-13.
Strengths and weaknesses
Reasoning, maths and science: strong, ahead of 78% of the field on average across 2 benchmarks. Its best board here is MMLU-Pro (#16 of 136) and its weakest is GPQA Diamond (#90 of 275).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so Step-3.5-Flash costs less than the typical listing. Models that score within 1.5 points or better and cost less: DeepSeek V4 Flash 0731 (max) ($0.016, 53.1), GLM-5.2 (high) ($0.060, 51.6), Qwen3-Coder-30B-A3B-Instruct ($0.070, 52.7). Similar scores cost more on Qwen3.5-122B-A10B ($0.113) and MiMo-V2.6 Flash ($0.140).
Reading these numbers
The 2 scores on this page come from 2 sources: HuggingFace GPQA Diamond leaderboard (1), HuggingFace MMLU-Pro leaderboard (1). No official leaderboard supplied any of them, so treat them as third-party measurements. With 2 results the aggregate rank can move a lot when one new score arrives.
Models ranked nearby
- #61 Claude Code + Fable 5.1 (52.6)
- #62 claude-opus-4-20250514 (52.5)
- #63 Claude Opus 4.6 (max) (52.5)
- #65 Tencent Hy4 preview (unknown) (52.4)
- #66 Muse Spark 1.1 (xhigh) (52.4)
- #67 DeepSeek-V3.2-Exp (high) (52.4)
Compare it on the live leaderboard → · How the aggregate score works