A Tour of Every Model in Lava

It is a great time to be using AI. More labs are shipping more powerful models than ever, each with its own strengths and its own personality. It seems every month a new model tops a benchmark, cuts a price, or unlocks a kind of work that was out of reach before.

Lava lets you use the apps you love with the AI you want. The model picker in Lava Desktop carries models from Anthropic, OpenAI, Google, xAI, Meta, and the leading open-source labs, and as new models come out we keep adding the best of them. Switching is one click, so you can experiment freely and find the model that fits the work in front of you.

Below is a tour of every model currently in Lava, what each one is best at, and how we think about choosing.

Key Takeaways

• When in doubt, pick Default. We keep it pointed at the best combination of intelligence, speed, and cost for most Lava work, and we update it automatically as the field moves.
• Match the model to the job. Fast and cheap for daily tasks, frontier reasoning for hard ambiguous problems, ultra-cheap for high-volume routines.
• Switching models is the point. You are one click away from a different brain. Experimenting is cheap, so try a few and see what fits your work.
• Tell us what is missing. If there is a model you want in Lava, we want to hear about it.

Start With Default

Lava comes preloaded with a Default model. We set it based on all of our research and data: the best combination of intelligence, speed, and cost for most work done in Lava. As new models come out and as our data gets better, we update Default automatically. When you pick Default you are always getting the current Lava recommendation, without having to keep up with model news yourself.

For everyone who wants to go deeper, here is how we think about the rest of the lineup.

The Daily Drivers

Most knowledge work does not need frontier-heavy reasoning. It needs a model that is fast, smart enough, and cheap enough that you never hesitate to use it. These two are built for exactly that.

GLM 5.2

Z.ai's open-source workhorse, and my pick for the high-speed daily grind. Best for fast, iterative agent tasks, multi-step workflows, and general knowledge work. It is blazing fast, with faster peak speeds than anything in Claude's lineup, and intelligence approaching Claude Opus 4.8. For rapid, cost-effective day-to-day execution, GLM 5.2 is the sweet spot.

Gemini 3.7 Flash

Google's fast coder and strategic executor. Best for code execution, data manipulation, UI and design tasks, and long high-context agent runs. It gets near Claude Opus 5 intelligence while running at near-GLM speeds, and it likes to batch its work and execute through code, which makes it extremely efficient. Slightly cheaper per task than GLM 5.2 and significantly cheaper than Opus.

The Deep Thinkers

When the problem is genuinely hard, ambiguous, or convoluted, raw intelligence starts to matter more than speed or budget. These are the models I reach for when I want the best answer, not the fastest one.

Claude Opus 5

Anthropic's flagship and the smartest model in the picker. Best for intricate workflow planning, deep conceptual extraction, complex architectural reasoning, and problems you cannot quite articulate yet. It is also the slowest and most expensive option here, so use it when intelligence is the thing you are paying for.

Claude Sonnet 5

Anthropic's polished generalist. Best for nuanced writing, thoughtful synthesis, and well-rounded reasoning. It is very capable and refined, though generally slower and pricier than GLM 5.2 or Gemini 3.7 Flash. A great pick if you value Anthropic's distinct prose and steady execution style.

ChatGPT 5.6 Sol

OpenAI's frontier intelligence at high speed. Best for complex tasks that need frontier-level reasoning without the latency penalty of the heavier flagships. Sol delivers top-tier thinking at a cheaper price point than Opus 5 and with significantly faster generation.

Kimi K3

Moonshot's open-source long-context specialist. Best for heavy document analysis, long-context reasoning, and complex extraction. Frontier-level intelligence at pricing below ChatGPT 5.6 Sol, closer to Claude Sonnet 5's tier.

DeepSeek V4 Pro

DeepSeek's open-source technical model. Best for deep technical generation, backend logic, and structured data tasks. It has an outstanding price-to-performance ratio for logic-heavy agent work.

The Specialists

Some models are trained for a specific shape of work rather than general brilliance. When your task matches their specialty, they punch above their weight.

Muse Spark 1.2

Meta's dedicated tool-calling model. Best for structured agent workflows, deterministic tool invocation, and API execution. It sits between GLM 5.2 and Gemini 3.7 Flash on intelligence, runs slightly slower than both, and is priced at their level. When the job is "call these tools in the right order, every time," Muse Spark is built for it.

Grok 4.6

xAI's cost-effective agentic reasoner. Best for general agent tasks, factual queries, and step-by-step tool work. Strong cost-per-task efficiency on par with Gemini and GLM, with moderate throughput.

The Budget Picks

Some work runs all day, every day. For that, the per-task price is the feature.

ChatGPT 5.6 Luna

OpenAI's ultra-low-cost automation model. Best for repetitive, well-defined scheduled routines, simple extraction, and background tasks. Luna is incredibly cheap: you can run high-frequency tasks all day without draining your balance. It is fast, but less robust on long, ambiguous, multi-branch work, where it can lose the thread.

ChatGPT 5.6 Terra

OpenAI's mid-tier model. We are still figuring out where it fits. It is faster than Sol but less intelligent, and significantly pricier than Luna. We have not found a clear job for it between Luna and the daily drivers, and we are evaluating whether to retire it. If Terra is your favorite, tell us why.

Try Them All

Please try the different models. Switching is one click in the composer, and running the same task through a second model is the quickest way to find the right fit for your work.

At Lava we will always work hard to bring you a wide range of AI model choices. If there is a model you would like added, or you have questions about picking the right one for your work, let us know at support@lava.so. And if you have not tried any of them yet, download Lava and start with Default.