Blog A coding gateway that cuts enterprise AI coding costs in half at frontier performance.
Today, Via is launching out of stealth to provide enterprises frontier AI coding at half the cost. We’re already in trials at several of the largest tech companies in Silicon Valley and are in production at a public healthcare company.
We’re a team of researchers, Olympiad camp qualifiers (with multiple international gold medals), and repeat founders from Stanford and IIT. Our backers include Jeff Wilke (ex-Amazon.com CEO), Manik Gupta (ex-Uber CPO), and former VPs at Anthropic, Google, Apple, Snowflake, and Twitter.
Exponentiating token spend is one of the largest problems enterprises face today, and AI coding constitutes the majority of it. By 2028, Gartner projects AI coding costs will exceed the average developer’s salary. Existing solutions like routing compromise coding accuracy and disrupt the developer experience.
Via’s solution is not a router, nor a harness, nor an observability layer. It’s a middleware that sits between a CLI / IDE and an LLM, installs in 10 seconds, and has already demonstrated a new standard of performance:
- On DeepSWE 1.1, Via achieves a 67.4% cost-reduction and is equal in accuracy to Claude Code.
- On Terminal Bench 4, Via achieves a 33.5% cost-reduction and is more accurate than Claude Code.
Via’s approach stems from Codegraph, the LLM-optimized compression algorithm we built out of Stanford CS. Through Codegraph, Via led SWE-Bench Verified for 5 months of 2025 as the most accurate coding harness, ahead of Anthropic, OpenAI and others (single model, no test-time compute).
What it means for your team: #
Integration:
- Your engineers keep the CLI / IDE they love. Setup is one terminal command.
- Works with public and internal gateways.
Security:
- Via runs inside your VPC, so no data leaves your environment.
- PII/PHI redaction, agentic threat detection, and optional on-prem inference.
- We provide two configurations: Via and Via US. Via US uses only American models.
Finance:
- Role-based budgets, spend observability, and forecasting
- Save millions on token costs without complaints from your developers!
Via is designed for enterprises with seven-figure coding budgets and above. We are taking on five more customers this year, and we’ll work hands-on with each one ourselves. Email pilot@via.dev and we’ll get back to you within 24 hours.