Step 3.5 Flash 2603 Pricing, Benchmarks & Channels

Step 3.5 Flash 2603 is a 2026-04-02 model from StepFun featuring a 256K context window and extended reasoning capabilities. Lowest rate on StepFun Official (CN) ($0.1046 in / $0.3138 out per 1M tokens).

⚖️ Compare in Radar Workbench 🧮 Monthly Cost Simulator
Input Price / 1M
$0.10461
Output Price / 1M
$0.31384
Cache Read / 1M
$0.02092
Context Window
256K
Generation Speed
141 TPS
Access Grade
A · CN Direct

F.A.R.E. Four-Dimension Score Breakdown

A · Accurate (Capability) 65/100 · Elo 1387 F · Fast (Speed & Latency) 141 Token/s · TTFT 1749ms R · Reliable (Stability & Access) Grade A · CN Domestic / CN Direct E · Economical (Cost) $0.10461 in · $0.31384 out
Scenario Capability Breakdown (0–100):
💻 Coding
67/100
✍️ Writing
58/100
📚 Long-Doc RAG
59/100
👁️ Vision OCR
Text-only
F.A.R.E. CAPABILITY RADAR
A · Accurate F · Fast E · Economical R · Reachable

Step 3.5 Flash 2603 Multi-Platform Price Comparison (1 Verified Channels)

All prices are in USD per 1M tokens. Domestic developers can prioritize channels marked "Direct in CN" for compliant access and local billing.

Platform / Gateway Network Access Input ($/1M) Output ($/1M) Cache Hit ($/1M) Billing & Notes Link
StepFun Official (CN) Lowest Price Direct in CN $0.1046 $0.3138 $0.0209 Direct CN access · RMB billing Visit Official →

✅ Core Strengths & Ideal Workloads

  • Fast generation speed (~141 tokens/sec)
  • Ultra-low token cost, ideal for high-volume batch workloads
  • Direct access in mainland China with RMB enterprise billing
  • Open weights available for private self-hosting (~20GB VRAM at INT4)

⚠️ Limitations & Practical Caveats

  • Domestic Chinese frontier lab specializing in multimodal reasoning and fast flash models.
  • Extended thinking generates internal reasoning tokens (~1.8x output multiplier)