New! Step-3.7-Flash is now available
tiyuvta inference
Menu
model · live

Step-3.7-Flash

A 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model that combines a 196B-parameter language backbone with a 1.8B-parameter vision encoder for native image understanding. Engineered for high-frequency production workloads, it activates approximately 11B parameters per token.

StepFun’s own description of the model, from Step-3.7-Flash model card . Served here as stepfun/step-3.7-flash.

StepFun mark
StepFun

What StepFun built it for

in the vendor’s words

Quoted from Step-3.7-Flash model card, retrieved 29 Aug 2026.

Published benchmarks

StepFun’s figures, not ours
Step-3.7-Flash: scores as published by StepFun, retrieved 29 Aug 2026
Benchmark What it measures Score How StepFun ran it
ClawEval-1.1 Agent execution reliability 67.1 The card states it leads this benchmark, next closest competitor at 59.8. Mode: not stated for this row.
SWE-Bench PRO Agentic coding, real-world software engineering 56.3 The card states a second-place finish. Mode: not stated for this row.
Terminal-Bench 2.1 Agentic terminal coding 59.5 The card groups this with rows it calls areas for future optimization. Mode: not stated for this row.
Toolathlon Tool use 49.5 as published. Mode: not stated for this row.
HLE w. Tool Reasoning with tools 48.1 as published. Mode: not stated for this row.

Every score above is StepFun’s own published result for this model, copied from Step-3.7-Flash model card on 29 Aug 2026. We did not run these benchmarks and we do not restate them as our own. Each vendor ran its own harness with its own judge, so these tables are not a league table: scores from two different cards are not comparable, even where the benchmark name matches. Each model here shows only its own maker’s table, for that reason.

Recommended usage

what StepFun recommends

Sampling

For recommended inference settings - the default path, and what we serve, StepFun recommends temperature=0.5, top_p=0.9. That is what this endpoint sends when your request omits them, so the default request shape is the vendor’s recommendation rather than ours. An explicit value in your request always wins.

Thinking

Thinking is on by default and streams before the answer; the card offers three selectable reasoning levels (low, medium, and high) so developers can balance speed, cost, and cognitive depth.

Context length

StepFun gives the native window as 262,144 tokens (the card describes it as a 256k context window), which is what we serve: the full checkpoint window, with input and output sharing it. One shared window for input and output: an omitted max_tokens is bounded by the remaining context.

What to point it at

StepFun names these workloads for this model:

  • “parsing massive financial reports in one pass, running multi-step search loops with cross-source verification, or operating concurrent coding agents in high-throughput pipelines”

Recommendations on this page are StepFun’s, from Step-3.7-Flash model card, retrieved 29 Aug 2026. We default to them rather than to a house preference; where our default would differ, this page says so.

What it does on this endpoint

our measurements · not vendor figures
output speed up to 140 tok/s single stream, measured live
context 262K tokens, the native window

These three are ours, measured on the live endpoint: observations, not service guarantees, and not to be confused with the vendor scores above. The protocol, the vantage, the build each figure is pinned to and how to check it against the system_fingerprint in your own response are all on the measurements page. Run your own and use those instead.

Price

per million tokens · no request fee · no subscription
Token rates: Step-3.7-Flash
Input Cached input Output
$0.20$0.03$1.00

One price, no tiers. Cache hits are applied automatically and reported per response as cached_tokens. Full price sheet and worked examples.

Capabilities

stepfun/step-3.7-flash

Three API formats, one base URL

OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages, all served from https://api.tiyuvta.ai/v1. Point an existing client at it and set the model id; there is no product-specific SDK.

Tools and structured output

tools and tool_choice, JSON mode, and JSON schema via response_format. The tool-call round trip is part of the release gate on every deploy.

Vision

Images go in as image_url content parts and bill as ordinary input tokens; vision costs nothing extra.

Prompt caching

Continuing a conversation reuses its prefix and bills the cached rate for what actually hit, reported per response, priced per token, no hit-rate promises.

Exactness-gated serving

Speculative decoding is verified byte-identical to plain decode on the serving configuration before a build takes traffic; speed never changes the output. The correctness gates.

First request

stepfun/step-3.7-flash
shell · default https://api.tiyuvta.ai/v1 
curl https://api.tiyuvta.ai/v1/chat/completions \
  -H "Authorization: Bearer $TIYUVTA_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "stepfun/step-3.7-flash",
       "messages": [{"role": "user", "content": "Hello"}]}'

The default request reasons before it answers, at StepFun’s recommended sampling: best quality for hard problems.

Get an API key Agent client configurations Read the docs
Weights and vendor sources
huggingface.co/stepfun-ai/Step-3.7-Flash-NVFP4, Apache-2.0. StepFun trained this model; we serve it at its full native context in quantized precision. Benchmark scores and usage recommendations on this page come from Step-3.7-Flash model card , retrieved 29 Aug 2026.
Machine-readable metadata
GET https://api.tiyuvta.ai/v1/models for supported parameters, and the provider feed at GET https://api.tiyuvta.ai/models?schema=openrouter.
Rate limits
Per account, for abuse control, raised on request; capacity scales with demand rather than sitting behind a published ceiling.