Step 3.7 Flash
Step 3.7 Flash
An Apache-licensed vision model built for coding agents, and it routes work rather than brute-forcing it.
Updated
The read
Step 3.7 Flash is a sparse vision-language model aimed squarely at coding agents and search workflows, with a routing mode that sends easy sub-tasks to cheap paths and hard ones to expensive paths. On coding tasks it lands close to models costing many times more per run. Apache-licensed, which for a model this specialised is unusually generous. The context window is modest by current standards.
Where it shines
- Apache 2.0 on a model tuned specifically for agents
- Routing keeps the cost per completed task very low
- Reads screenshots and diagrams, not just text
Worth knowing first
- Context window is modest next to the million-token field
- Narrow: general chat is not what it is for
Index score
81.6
Provisional. This score is read from the product's public capability, not from the one-prompt rebuild the fully reviewed entries went through, so treat it as a placing rather than a verdict.
- Rank
- 18 of 53
- Builds
- LLMs
- Category
- Coding
- Output
- Apache-licensed weights, plus a cheap API
- Pricing
- Free tier
- Best for
- Coding agents that need to watch their own cost per task
Scores are our own editorial judgement, weighted craft 55, speed 10, control 15 and value 20.
Also worth a look
Full index- 04Claude Sonnet 5Claude Sonnet 5ProvisionalThe coding workhorse: near the top on real software tasks, at a working price.Coding modelPaid only87.9
- 10MiniMax M3MiniMax M3ProvisionalOpen weights tuned for software work, and the fastest decoding of anything in its class.Coding modelFree tier84.5
- 43Mercury 2Mercury 2ProvisionalA diffusion model for text, and it is several times faster than anything built the usual way.Coding modelPaid only73.2
