Step 3.5 Flash
by Stepfun · stepfun/step-3.5-flash · #45 cheapest of 325 paid models
Prices updated Jul 22, 2026, 1:31 AM UTC · refreshed hourly
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....
Compare vs:Claude Sonnet 5Claude Opus 4.6Claude Haiku 4.5GPT-5.6 Sol
Pricing
Input
$0.10
per 1M tokens
Output
$0.30
per 1M tokens
Blended (3:1)
$0.15
3 input : 1 output
Cache read
—
per 1M cached tokens
Cache write
—
per 1M tokens
Cache & batch economics
Effective prices for Step 3.5 Flash (no cache-read rate published — hits billed as normal input).
| Cache hit rate | Effective blended $/1M | Example request* | vs no cache |
|---|---|---|---|
| 0% | $0.15 | $0.0105 | — |
| 50% | $0.15 | $0.0105 | −0% |
| 90% | $0.15 | $0.0105 | −0% |
*Example: 100K sticky context + 2K new input + 1K output. Batch discount is an approximate 50% off list rates (available for OpenAI / Anthropic / Google — not flagged for this creator). Confirm on the vendor’s pricing page.
What a request costs
| Workload | Input tokens | Output tokens | Cost / request | Cost / 1K requests |
|---|---|---|---|---|
| Short chat message | 500 | 300 | $0.00014 | $0.14 |
| Document summary | 8,000 | 1,000 | $0.0011 | $1.10 |
| Codebase question (RAG) | 30,000 | 2,000 | $0.0036 | $3.60 |
| Long-context analysis | 150,000 | 5,000 | $0.0165 | $16.50 |
Performance
Intelligence Index
26
AA composite quality
Coding Index
—
Math Index
—
Agentic Index
—
Output speed
298 tok/s
median
Time to first token
0.74s
median
Time to first answer
7.45s
after reasoning tokens
Artificial Analysis benchmarks
| Benchmark | Score |
|---|---|
| GPQA Diamond | 82.6% |
| Humanity's Last Exam | 22.6% |
| SciCode | 38.5% |
| IFBench | 66.5% |
| AA-LCR (Long Context Reasoning) | 54.3% |
| Terminal-Bench Hard | 32.6% |
| τ²-Bench (Telecom) | 87.4% |
Specs
Context window
262K
tokens
Max output
66K
tokens
Input modalities
text
Tokenizer
Other