Skip to content
OSMEXO

Step 3.7 Flash

stepfun/step-3.7-flash
  • StepFun
  • Family · Step 3 Flash
  • Released 29 May 2026
  • Open weights
  • Text · Image · Video → Text
  • Reasoning · adjustable
  • Tools · JSON · Vision
Not on OSMEXO yet
Spec & list price
Context256K
Max output225K
Input / 1M$0.20
Output / 1M$1.155.7× input
Scores · standing
Intelligence19.5#145/489
Coding39.6#90/196
Usage
Tokens / day79.6B▲ 234% 7d
Usage rank#38
Requests / day890.5K
Tool calls / req0.57
Per request · 30 days
Avg input58.0K3.9× median
Avg output1.5K2.1× median
Cache hit92%1.3× median
Tool-call failures2.4%1.2× median

Consumption

26.7B53.3B80.0B3 Oct17 Oct1 Nov
Peak 75.2B 28 OctLow 1.2B 11 OctRank today #38

Benchmarks

bar = share of the top score · Benchmark data: Artificial Analysis
τ²-Bench#4 of 319
98.5%
GPQA Diamond#129 of 445
80.9%
Long-context reasoning#111 of 393
73.7%
IFBench#75 of 328
67.3%
SciCode#89 of 160
43.9%
Terminal-Bench Hard#50 of 311
35.6%
Coding index#90 of 196
39.6
Terminal-Bench 2.1#93 of 182
39.3%
Speed · median across 2 providers, latest day
Output speed#277/33438 tok/s1.7× slower than median
First token#3154.81 s4.7× longer than median

#7 cheapest of 24 with a similar score

Value
Intelligence Index
1618202224$0.1$0.2$0.5$1Step 3.7 Flash
Blended price per 1M tokens (log)
Family progress
Intelligence Index
17181920MarAprMayStep 3.7 Flash
  • Family releases19.5

Technical details

Reasoning
on · medium
Tokenizer
Other
License
Apache 2.0
Parameters
201B
Moderation
none
Downloads · likes
12.0K · 462
Reasoning effort
  • high
  • medium · default
  • low
Request parameters 18
  • tools
  • tool_choice
  • response_format
  • structured_outputs
  • reasoning
  • reasoning_effort
  • include_reasoning
  • temperature
  • top_p
  • top_k
+8 more
  • max_tokens
  • stop
  • seed
  • frequency_penalty
  • logprobs
  • presence_penalty
  • repetition_penalty
  • top_logprobs

Interesting facts

  1. 2.2× the output room of similar models (230.4K against a 106.5K median).
  2. Released 157 days ago, the newest of 2 Step 3 Flash models.
  3. Weekdays average 23.6B tokens a day against 25.1B on weekends.

When to use

Best for

  • Tool-calling agents on a budget
  • Image and video understanding
  • Long-context question answering
  • Reasoning with adjustable effort at lower cost

Think twice for

  • Terminal-heavy coding agents
  • Hardest coding benchmarks
  • Image or audio generation

Same family

Step 3 Flash

Alternatives

other vendors · similar capability

Want Step 3.7 Flash on OSMEXO?

Sign in to request it. Requests are reviewed; a request does not guarantee availability.