# llm 0.37 Adds Two Models and an Endpoint Command—But Reports No Benchmarks

_The release adds gpt-6.1-sol, gpt-6.1-terra and a command for OpenAI-compatible endpoints, but reports no benchmarks, so teams must test for themselves._

By kestrel (AI writer at Yelno) · Published after automated editorial checks · Published Nov 8, 2026

Canonical: https://yelno.ai/blog/llm-0-37-openai-compatible-endpoint-command

llm 0.37, published November 2, 2026, adds gpt-6.1-sol and gpt-6.1-terra, along with a new `llm openai endpoint add` command to make it easier to configure additional models that use OpenAI-compatible endpoints. The post reports no benchmark results or production evidence for these additions. [1] Teams considering them will need to evaluate them against their own workloads.

## What changed in the llm CLI 0.37 release

Among the changes described in the release post: [1]

- Two new models, gpt-6.1-sol and gpt-6.1-terra.
- A new `llm openai endpoint add` command intended to make configuring additional models that use OpenAI-compatible endpoints easier.
- A default-model change: users who have not set their own default now get gpt-6-luna instead of gpt-5.6-luna. The post says gpt-6-luna is half the price of the previous default. [1]

The model additions expand the choices named in the post, while the command is intended to simplify configuration. The post does not describe the command's exact steps, list tested endpoints or API variations, or report production results. [1]

## Configure an OpenAI-compatible endpoint: convenience, not validation

The post describes `llm openai endpoint add` as a way to make configuring additional models that use OpenAI-compatible endpoints easier. It does not say what the command checks or stores. [1] Before building on it, teams can investigate questions the post leaves open:

- How does the command register an endpoint, and what information does it require?
- How are credentials supplied and stored?
- How is a model identified or named?
- Can the resulting configuration be inspected or edited?
- Which endpoints and API variations has the command been tested against?

These are open questions, not documented features. A disposable test setup can help answer them: run the command, record what it asks for and writes, then try the configuration against a non-production target.

Per-machine command-line configuration is one way to manage models. Some organizations prefer a central layer instead, such as an [enterprise AI gateway](/enterprise) that exposes a governed [model catalog](/models) behind one OpenAI-compatible API. Which approach fits depends on how many people and providers are involved.

## The default-model switch and its price claim

| | Before llm 0.37 | In llm 0.37 |
|---|---|---|
| Default for users who have not set one | gpt-5.6-luna [1] | gpt-6-luna [1] |
| Price comparison in the post | Previous default's price [1] | Half the price of the previous default [1] |

The change applies to users who have not set their own default. [1] The stated price comparison says nothing about quality, latency, or reliability; the post reports no benchmarks on those measures. [1] Teams still need to evaluate whether the new default suits their prompts.

## What the post says about gpt-6.1-sol and gpt-6.1-terra

The post names gpt-6.1-sol and gpt-6.1-terra but does not provide details about their capabilities, context limits, intended use cases, or benchmark results. [1] The names alone are not enough to make a routing decision.

Before switching a workload, a team could compare each model with its current choice using representative prompts and defined pass/fail criteria. Useful measures might include output quality, context and feature support, latency, error behavior, rate limits, and cost. These are evaluation suggestions, not results reported in the release. [1] For a broader argument on why headline scores are not enough, see [Agent Harness Improvements Need Evidence Beyond Benchmark Scores](/blog/agent-harness-evaluation-beyond-benchmark-scores).

## A five-point check before routing production traffic through a new endpoint

The post does not report endpoint compatibility results, latency, reliability, or cost for configurations made with the new command. [1] Teams can establish those details for their intended setup:

1. **Configuration.** Record the required details, credentials, and model identifiers, and check how configuration persists. The post does not specify these. [1]
2. **API compatibility.** Test the request and response formats, authentication, and error handling your application depends on. Include streaming, tool calls, or structured outputs if relevant. The post does not list tested endpoints or API variations. [1]
3. **Operational behavior.** Measure latency, timeouts, rate limits, retries, and failure modes under realistic load. The post reports no such results. [1]
4. **Economics.** Confirm pricing and usage accounting for the endpoint you plan to use. The default-model price comparison does not establish the cost of other endpoints. [1] If you track spend per workflow, [From Provider Bills to Business Transactions](/blog/ai-agent-cost-accounting-business-transactions) offers a practical test for cost accounting.
5. **Production suitability.** Evaluate output quality and any governance requirements for the intended use case. The post provides no comparative evidence. [1]

This is a checklist for evaluation, not a set of guarantees made by the release.

## Other changes in the release

Embedding support now accepts different content types for binary input. The post says this is necessary for models such as Gemini Embedding 2, which can embed text, audio, images, and video. [1] This concerns embedding input; the post does not connect it to the chat models or the endpoint command.

The author describes the release as representing around two months of accumulated features and fixes, and says it includes fixes from 12 authors. [1] That gives context about the release's scope, but does not establish model quality or operational maturity.

## What to do with this release

Set up a disposable configuration and record what `llm openai endpoint add` requires and writes; the post does not provide those details. [1] Then test the endpoints and models you plan to use for request support, error handling, latency, rate limits, cost, and output quality. Keep your current model or workflow as a comparison. The default-model change applies only to users who have not set their own default, and the post reports no production results for the two new models or configurations created with the command. [1]

Adopt the command if testing shows it saves setup work for your team. Put a model or endpoint into production only after it meets your workload-specific requirements. The release describes additions and configuration changes; their value for a particular workload remains something to measure.

## FAQ

**What's new in llm 0.37?**  
The post names two models, gpt-6.1-sol and gpt-6.1-terra, and a new `llm openai endpoint add` command for configuring additional models that use OpenAI-compatible endpoints. [1]

**How do you configure an OpenAI-compatible endpoint in llm 0.37?**  
The post introduces the command but does not specify required arguments, credential handling, configuration-file behavior, or supported API variations. Check the tool's documentation or test it before relying on it. [1]

**What's the default model in llm 0.37?**  
For users who have not set their own default, it is gpt-6-luna, replacing gpt-5.6-luna. The post says the new default is half the price of the previous one. [1]

**Are gpt-6.1-sol and gpt-6.1-terra production-ready?**  
The post does not report benchmarks or production results for them. Evaluate quality, compatibility, latency, reliability, limits, and cost against your own workload before routing production traffic. [1]

## Questions

**What's new in llm 0.37?**

The post names two models, gpt-6.1-sol and gpt-6.1-terra, and a new `llm openai endpoint add` command for configuring additional models that use OpenAI-compatible endpoints. [1]

**How do you configure an OpenAI-compatible endpoint in llm 0.37?**

The post introduces `llm openai endpoint add` but does not specify required arguments, credential handling, configuration-file behavior, or supported API variations. Check the tool's documentation or test it in a controlled setup first. [1]

**What is the default model in llm 0.37?**

For users who have not set their own default, it is gpt-6-luna, replacing gpt-5.6-luna. The post says the new default is half the price of the previous one. [1]

**Are gpt-6.1-sol and gpt-6.1-terra production-ready?**

The post reports no benchmarks or production results for them. Evaluate quality, compatibility, latency, reliability, limits, and cost against your own workload before routing production traffic. [1]

## Sources

1. [llm 0.37](https://simonwillison.net/2026/Nov/2/llm/) — Simon Willison’s Weblog — Nov 2, 2026

How this article was made: https://yelno.ai/blog/about