Fireworks Nexus: Routing Code Tasks to Open-Weight Models and Cutting AI Costs 3–5x

On July 28, 2026, Fireworks AI released Fireworks Nexus, a routing and cost-control platform that sends routine coding tasks to open-weight models. The developer says it cuts costs 3–5x without changing familiar tools such as Claude Code.

Why this was needed

Spending on AI agents is growing faster than companies can plan for it. According to Forbes, Uber used up its entire 2026 AI budget in just four months. Fireworks cites the same report: adoption of agentic tools in engineering teams rose from roughly a third of employees to more than four-fifths in two months, and Claude Code alone reached about 5,000 engineers after its launch in December 2025.

Fireworks describes the root of the problem as a mismatch rather than overspending: most organizations run routine work at "frontier" prices because switching to open-weight models is operationally inconvenient for platform teams.

How Fireworks Nexus works

Fireworks Nexus has three parts and plugs in on top of the development tools a team already uses. Key features:

• Control and observability — team- or company-level budgets, per-model ROI tracking, US-hosted endpoints, zero data retention, and coverage across 20 global data centers.
• FireConnect — a one-line install under the Apache 2.0 license that maps the model slots in a harness to Fireworks models. Claude Code, Codex and OpenCode keep working unchanged through Anthropic- and OpenAI-compatible APIs.
• Smart routing — a trained model scores the complexity of each request. Routine work goes to a cheap open-weight model, while hard tasks go to your existing provider using your own key (which, according to Fireworks, is not stored on the server).

The router is still a research preview. Right now it routes between Claude Opus 5 and GLM-5.2, so the pass-through path requires an Anthropic key. The fully open configuration routes between Kimi K3 and GLM-5.2.

How much cheaper is it?

The savings in independent tests were noticeable and cannot be explained by caching alone. Faros AI ran 211 real engineering tasks from 12 repositories across seven model-plus-harness routes. Claude Code on GLM-5.2 scored 0.568 on a model-judge rubric, against 0.521 for Claude Code on Opus 4.8, at a cost of $0.92 versus $1.76 per task. Cache share was 89.7% and 99.7%, respectively.

"This cohort is specific to one company and is not a universal leaderboard,"

Faros AI cautions in its summary of the measurement.

A second study, by Arize together with Fireworks, covered 10 models on 40 Terminal-Bench tasks with six runs each — 2,400 runs and $626 in API spend. On easy tasks, frontier models show no advantage: Kimi K2.6 passed 73% and GPT-5.5 passed 69%. On hard tasks the top models lead: GPT-5.5 at 51% and Kimi K3 at 32%. Simulated routing beat every single-model strategy. An escalation ladder cost $0.525 per successful task and consistently solved 32.3 of 40, while GPT-5.5 alone cost $0.636 and solved 25 of 40.

What this means

Fireworks Nexus turns model choice into a controllable parameter: routine work goes to cheap open-weight engines, and the expensive frontier model is reserved for genuinely hard tasks. The numbers come from the vendor and from its preview program, but the independent runs by Faros AI and Arize support the main thesis: on some tasks, overpaying for a top-tier model does not buy better quality.

Frequently asked questions

Does Fireworks Nexus work with Claude Code?

Yes. The FireConnect component installs with a single command from the Fireworks dashboard and maps model slots, so Claude Code, Codex and OpenCode keep working unchanged through Anthropic- and OpenAI-compatible APIs.

How much does Fireworks Nexus cut costs?

According to Fireworks, the typical reduction is 3–5x. In tests with the Notion and Doximity teams, the cost of a single merged pull request fell by roughly a third, and the blended token price came out at about a quarter of that of closed model labs.