What mammoth-1 is
mammoth-1 is a decision model from Base Labs, the research team at Baseten, the AI inference company. Baseten released it on 21 October 2026 together with a 9B sibling, elephant-1. Base Labs post-trained Alibaba’s Qwen3.5-27B on 6.21 million decision examples, with a calibration term in the loss and a trained “I don’t know” option.
You send a state, text with optional images, and a set of typed questions: Choice, Score and Noul. It returns a probability for every option and writes no text. All the questions share one pass over the state, so each extra question costs a pass over the question alone.
Access and price
Baseten serves it at POST https://inference.baseten.co/v1/systemone, the same request shape as Jev, with the model id decision-models/baseten-mammoth-27b-latest. It is free for one week on Baseten’s Model APIs, until 27 October 2026, and OpenRouter lists a free version served by Baseten. Baseten’s pricing page lists the model without a price, so the cost after the free week is not published yet. A request can hold 16,384 tokens; longer ones are refused.
The weights are on Hugging Face under MIT. The card’s licence changed from a proprietary, evaluation-only licence to MIT on 21 October, the day of the launch.
What the benchmarks show
On the Jev Decision Index 0.4.1, the file generated on 22 October 2026, mammoth-1 scores 68.44 on the Full score, 3rd of 229 rows, against 60.24 for Jev 1.14. Its public score of 75.34 is the highest on the board, but Base Labs submitted that run. Its calibration error, the average gap between stated confidence and how often it is right, is 0.0408 against Jev 1.14’s 0.0799.
Baseten’s blog and card say mammoth-1 is first of 208 models at 71.24. That figure comes from the board’s earlier 0.4 edition. In 0.4.1 it sits behind Fastino’s GLiDE at 71.52 and Jeffan-31b-1710 at 69.75.
Base Labs reports 65.9 on its own replica of the Artificial Analysis Decision Index. Artificial Analysis has not listed mammoth-1 itself.
Things to check
The technical report says Base Labs removed training examples that overlapped either Decision Index. It also says the team measured each data mix on the Jev Decision Index while choosing it. The Index still found signs of overlap with one private new-domain test and left that test out of the model’s score.
The Index measured a median of 220 ms per request on one RTX PRO 6000, slower than elephant-1’s 114 ms. Baseten publishes no latency for its hosted API.
Specifications
| System One Score | 95.7 from 1 of 3 boardsJev Decision Index 95.7. How the score works |
| Question types | ChoiceScoreNoul |
| Max Choice options | Not documented |
| Score levels | Not documented |
| Questions per call | Not documented |
| Total context | 16,384 tokens |
| State budget | Not documented |
| Rate limit | Baseten publishes no rate limit for the model as of 2026-10-23. |
| Endpoint | POST https://inference.baseten.co/v1/systemone |
| Licence | MIT |
| SDKs |
The model card gives a maximum sequence length of 16,384 tokens and says longer requests are refused, not truncated. OpenRouter lists a 16,384-token context. The card and Baseten's library page say it reads text and images; OpenRouter lists text input only. Self-hosting needs one CUDA GPU and the pinned runtime that setup.sh installs.
Versions
- decision-models/baseten-mammoth-27b-latest, 21 Oct 2026, The model id on Baseten's Model APIs. OpenRouter lists it as baseten/mammoth-1:free, snapshot baseten/mammoth-1-20261021. Weights on Hugging Face as baselabs/mammoth-1, a full fine-tune of Qwen3.5-27B with 27.4B parameters. The Jev Decision Index scored revision 9d1962f. Release notes
Use cases
What people use mammoth-1 for, one page per pattern.
Support inbox triage with System One models
Send a support ticket to Jev once with every question attached. Category comes back as a selected label, severity and frustration as numbers on scales you wrote, refund intent as a probability. Your code reads those values and decides what happens to the ticket.
Intent and model routing with System One models
One Jev call reads an incoming request and returns its intent as a label plus a difficulty rating on a scale you wrote. Your router reads both numbers and picks the handler: deterministic code, a cheap model, an expensive one, or a human queue.
LLM guardrails with System One models
Put one Jev request in front of an LLM and one behind it. Yes/no questions return the probability that each hazard holds, a Score rates how much harm complying would do, and your thresholds turn those numbers into pass, review, block, or a crisis path.
Confidence-gated actions with System One models
Jev returns a confidence value from 0 to 1 alongside every Choice and Score answer. Your code treats it as a separate axis: act automatically when it's high, confirm or flag when it's middling, hand the decision to a person when it's low. Riskier actions get higher bars.
Benchmarks
Published scores, each linked to its source. A row marked vendor-run was measured by the vendor itself, so read it as one side's result. The compare hub sets these beside the other models.
| Benchmark | Metric | Score | Run by | Date |
|---|---|---|---|---|
| Jev Decision Index 0.4.1 | Full score | 68.44 score | Independent | 22 Oct 2026 |
| Jev Decision Index 0.4.1 | Public score | 75.34 score | vendor-run | 22 Oct 2026 |
| Jev Decision Index 0.4.1 | Expected calibration error | 0.0408 ECE | Independent | 22 Oct 2026 |
| Artificial Analysis Decision Index (Base Labs replica) | Index score | 65.9 score | vendor-run | 21 Oct 2026 |
- Jev Decision Index, Full score: Row: baselabs mammoth-1 (27B), 3rd of 229 rows by this score. Counts 60% private same-skill tests and 40% private new-domain tests; public benchmarks no longer count. 0 is random guessing, 100 is perfect. The Index ran the private tests on one RTX PRO 6000 with Base Labs' serving code. It left out one private new-domain test for this model after finding signs that its training data overlapped it. Jev 1.14 scores 60.24.
- Jev Decision Index, Public score: Row: baselabs mammoth-1 (27B), the highest public score of 229 rows. Base Labs submitted the public run, and the Index rescored it with its own kit. Jev 1.14 scores 60.7.
- Jev Decision Index, Expected calibration error: Row: baselabs mammoth-1 (27B). Lower is better. Jev 1.14 scores 0.0799.
- Artificial Analysis Decision Index (Base Labs replica), Index score: Base Labs' own replica of the Artificial Analysis Decision Index, run on AA's prompt formats. Artificial Analysis itself lists elephant-1 but not mammoth-1 as of 2026-10-23.