Compare

GoModel vs TensorZero

TensorZero is a Rust LLMOps platform: a gateway plus observability, evals, and model optimization. Its repository was archived on 12 June 2026 and gets no more releases. GoModel is an actively developed Go gateway focused on routing, access control, and cost.

Get started with GoModel See the benchmark

Active

vs archived

TensorZero is read-only since 12 June 2026

6/6

API workloads served

vs 2/6 for TensorZero

23×

lower p50 latency

2.2 ms vs 50.0 ms

8.7×

lower p99 latency

6.9 ms vs 60.1 ms

The benchmark

A gateway sits on every request, so its overhead adds up. Same machine, same mock backend, one gateway at a time.

Median latency

p50, chat completions · lower is better

23× faster

GoModel2.2 ms

TensorZero50.0 ms

Tail latency

p99, chat completions · lower is better

8.7× lower

GoModel6.9 ms

TensorZero60.1 ms

Time to first token

p50, streaming chat · lower is better

TensorZero 3.8× faster

GoModel5.0 ms

TensorZero1.3 ms

Throughput

peak sustained · higher is better

TensorZero 1.1× more

GoModel4,498 req/s

TensorZero4,784 req/s

Memory

peak RAM · lower is better

1.4× less

GoModel74 MB

TensorZero107 MB

Cold start

container start to first request · lower is better

TensorZero 1.4× faster

GoModel0.8 s

TensorZero0.6 s

Docker image

compressed pull size · lower is better

4.3× smaller

GoModel21 MB

TensorZero88 MB

GoModel 0.1.98 vs TensorZero 2026.6.0 on an AWS c7i.large (2 vCPU, 4 GiB), run 20260928-043817 (28 September 2026). 20,000 requests per workload at concurrency 10, five randomized trials. API workloads served: GoModel 6/6, TensorZero 2/6. TensorZero wins three rows, and one cause explains the profile: its server leaves Nagle's algorithm on, so each non-streaming response on a reused connection waits about 40-50 ms for the client's delayed ACK. That idle stall, not compute (5.7% CPU), sets its p50 and p99. Its first streamed token arrives in 1.3 ms, but its streaming p99 is pinned near 60 ms, and its peak throughput grows with every connection the sweep opens (about 19 req/s each) rather than hitting a ceiling. It has no Responses or Anthropic Messages endpoint, so 4 of 6 workloads failed. Rerun it with one command.

Beyond speed

Different goals. TensorZero optimizes models from feedback; GoModel runs the shared gateway.

	GoModel	TensorZero
Project status	Active, releases weekly	Archived 12 June 2026, read-only
Purpose	Gateway and control plane	LLMOps: gateway, evals, optimization
Client APIs	OpenAI Chat, Responses, Anthropic Messages	Native /inference, OpenAI chat, embeddings
Infra for logs, keys, limits	None; SQLite built in	Postgres or ClickHouse, optional Valkey
Keys, budgets, rate limits	API keys, budgets, rate limits	API keys, token and cost limits
Fine-tuning, prompt optimization	Not offered	SFT, DPO, GEPA, DICL
Evals and A/B tests	Weighted routing only, no evals	Evals and adaptive A/B tests
License	MIT core + commercial Pro	Apache 2.0

Which one fits?

An honest split.

Pick GoModel if

You need a maintained gateway with security fixes.
Clients use OpenAI Responses or the Anthropic SDK, not only chat completions.
You want keys, budgets, and audit logs without running Postgres, ClickHouse, or Valkey.
You want one binary with a built-in, authenticated dashboard.

TensorZero still fits if

You fine-tune from production feedback (SFT, DPO, GEPA, DICL) and can maintain a fork.
You need built-in evals and adaptive A/B tests over prompt and model variants.
You already run it and need time to plan a migration.

Migrating is a base-URL change plus model names: TensorZero'stensorzero::model_name::... becomes a GoModel model or alias.

Try it against your own workload

Benchmarks are an argument; your traffic is the proof. GoModel speaks the OpenAI and Anthropic APIs natively, so pointing an existing app at it is a base-URL change.

1 · Run GoModel

docker run --rm -p 8080:8080 \
  -e LOG_FORMAT=text \
  -e OPENAI_API_KEY="your-openai-key" \
  enterpilot/gomodel

2 · Point your SDK at it

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8080/v1",
    api_key="your-gomodel-key",
)

Full setup, providers, and configuration live in the documentation. Questions? Ask on Discord or book a 30-minute call.