Available models
Posit Assistant lets you choose which large language model handles your request. Each model offers a different balance of capability, speed, and credit usage, and you select the active model in the Posit Assistant interface.
The following models may be available with your Posit AI Pass subscription. We add, update, and retire models as the field moves, so treat the model picker in Posit Assistant as the authoritative list.
Different models consume your Posit AI Pass usage credits at different rates. The rates below are expressed relative to Claude Sonnet 4.6, which serves as the baseline. For example, a request that would use 1% of your credits on Sonnet 4.6 would use roughly 0.33% on Haiku 4.5 and roughly 1.67% on Opus 4.6.
The Relative Usage Amount column shows the inverse: how much usage you get for the same credit spend. For example, Gemma 4 26B gives you roughly 10x as much usage as Sonnet 4.6 for the same credits.
| Model | Provider | Relative Credit Cost | Relative Usage Amount |
|---|---|---|---|
| Gemma 4 26B | Google (open weights) | 0.1x | 10x |
| Claude Haiku 4.5 | Anthropic | 0.33x | 3x |
| GLM 5.2 | Z.ai (open weights) | 0.38x | 2.6x |
| Claude Sonnet 5 | Anthropic | 0.67x | 1.5x |
| Claude Sonnet 4.6 | Anthropic | 1x (baseline) | 1x (baseline) |
| Kimi K3 | Moonshot AI (open weights) | 1x | 1x |
| Claude Opus 4.6 | Anthropic | 1.67x | 0.6x |
| Claude Opus 4.8 | Anthropic | 1.67x | 0.6x |
Open weights models
Posit serves Gemma 4 26B, GLM 5.2, and Kimi K3 through our agreement with Baseten, on GPU infrastructure that does not retain data. Posit does not store your conversation histories unless you opted in to data retention at sign-up. See your data preferences and the FAQ on Zero Data Retention.
GLM 5.2 and Kimi K3 share a few characteristics worth knowing about:
- Speed: both stream almost twice as many tokens per second as the Anthropic models, which gives them a qualitatively different feel in interactive use.
- Tokenization: these models tokenize text more efficiently, so the real cost savings are larger than the per-token difference alone implies.
- Context: both have context windows of 250K tokens or more, and Posit AI Pass does not charge separately for cache writes on them.
For more on how these two models fit into Posit AI Pass, see our post on Kimi K3 and GLM 5.2.
Gemma 4 26B
Gemma 4 26B is an open weights model from Google, available in Posit AI Pass at one-tenth the credit usage of Sonnet 4.6 and roughly one-third the credit usage of Haiku 4.5. It is the lowest-consumption option in Posit AI Pass.
Gemma 4 26B works well for:
- Basic data analysis
- Importing, tidying, and visualizing data
- Information retrieval and short factual questions
It is less reliable than the Claude models on longer, multi-step tasks. You might see a strong response on one turn and a misread on the next. We do not recommend Gemma 4 26B for long-running agentic coding or complex package development. When you use it, plan to steer it actively and review its code and output closely.
GLM 5.2
GLM 5.2 excels at agentic coding, at roughly a third of the credit usage of Sonnet 4.6, close to Haiku 4.5 rates.
Its main limitation for our audience is vision: GLM 5.2 cannot see plots. That makes it a weaker fit for exploratory data analysis and other visual work, where reading a chart is often part of the task. Reach for it on code-heavy, multi-step tasks where you want to keep credit usage low.
Kimi K3
Kimi K3 is the most capable open weights model in Posit AI Pass, and notably well-rounded compared to other open weights releases rather than strong in one narrow area. It draws credits at roughly the same rate as Sonnet 4.6.
Use it when you want frontier-level capability without the credit usage of the top Claude models.
Claude models
Posit AI Pass sends Claude requests to Anthropic under a Zero Data Retention agreement.
Claude Haiku 4.5
A fast, lightweight model. Good for quick questions, simple edits, short refactors, and follow-ups that do not need deep reasoning. The Claude option that draws credits the slowest.
Claude Sonnet 4.6
A capable general-purpose model for writing and editing R code, exploratory data analysis, building Quarto reports, and typical agentic coding tasks. Serves as the baseline for the credit rates above.
Claude Sonnet 5
The recommended default for most data science work. More capable than Sonnet 4.6 at roughly two-thirds the credit usage, so it is the better starting point unless you have a reason to pick something else.
Claude Opus 4.6
A high-capability model for complex architectural decisions, large refactors, and hard debugging. It draws credits fastest. Reserve it for tasks that justify the higher consumption.
Claude Opus 4.8
The most capable model in Posit AI Pass, at the same credit usage as Opus 4.6. Use it for the hardest problems: difficult debugging, large multi-file refactors, and work where Sonnet has already struggled.
Choosing a model
A practical pattern:
- Start with Sonnet 5 if you are not sure.
- Drop to Haiku 4.5 or Gemma 4 26B for shorter, simpler tasks to stretch your credits further.
- Step up to Opus 4.8 when you hit a problem that needs more reasoning, or where a smaller model has already failed.
- Try Kimi K3 or GLM 5.2 for agentic coding, where their speed shows most. Keep in mind that GLM 5.2 cannot see plots, so stay with a Claude model or Gemma 4 26B for visual data analysis.
For more on how model selection affects your monthly credits, see the FAQ on credit consumption.
Models behind Next Edit Suggestions
Next Edit Suggestions (NES) runs on specialized models hosted by Posit, not on the models above. NES usage is unlimited, is not metered against your credit balance, and is not configurable from Posit Assistant.