{"slug": "liquidai-s-d1-2-ultra-tops-decision-index-with-near-perfect-confidence", "title": "LiquidAI's d1.2-Ultra Tops Decision Index With Near-Perfect Confidence Calibration", "summary": "Liquid AI released d1.2 and d1.2-ultra, omnimodal decision models accepting text, image, and audio input, with d1.2-ultra scoring 75 on the Artificial Analysis Decision Index v2.1 — the highest listed result and two points above the previous leader. Liquid AI reports a calibration error of 0.8 points for d1.2-ultra and 1.2 for d1.2 across 22,505 questions from 13 benchmarks, versus 2.7 for Jev 1.14, 3.7 for Perplexity Decider v1.3, and 7.2 for GPT-6 Luna on the Decisions API. The models add five answer types (value read, span extraction, bounding boxes, number with range, multi-select), are priced at $0.02 and $0.042 per million input tokens with input-only billing on the Liquid API, and report 98% receipt read accuracy, 94% RefCOCO localization, and 95 to 97% on audio classification tasks.", "body_md": "# LiquidAI's d1.2-Ultra Tops Decision Index With Near-Perfect Confidence Calibration\n\nLiquid AI's d1.2 and d1.2-ultra top the Artificial Analysis Decision Index with audio input, five new structured output tasks, and near-perfect calibration.\n\n- Liquid AI released [d1.2 and d1.2-ultra](https://www.liquid.ai/blog/d1-2?utm_source=alphasignal) , omnimodal decision models with text, image, and audio input.\n- d1.2-ultra tops the [Artificial Analysis Decision Index v2.1](https://artificialanalysis.ai/evaluations/artificial-analysis-decision-index?utm_source=alphasignal) at 75, two points ahead of prior best.\n- Five new answer types: value read, span extraction, bounding boxes, number with range, multi-select.\n- Calibration error of 0.8 for ultra and 1.2 for d1.2, well ahead of competitors.\n- Pricing: $0.02 and $0.042 per million input tokens, input-only billing, available on [Liquid API](https://console.liquid.ai/?utm_source=alphasignal) .\n- 98% receipt read accuracy, 94% RefCOCO localization, 95 to 97% on audio classification tasks.\n\nLiquid AI has released d1.2 and d1.2-ultra, two decision models that accept text, images, or audio and return schema-constrained answers with probabilities. On version 2.1 of the independent [Decision Index](https://artificialanalysis.ai/evaluations/artificial-analysis-decision-index?utm_source=alphasignal), d1.2-ultra scored 75, the highest listed result and two points above the previous leader. The [release post](https://www.liquid.ai/blog/d1-2?utm_source=alphasignal) also introduces five answer types, audio input, and input-only usage pricing.\n\nTyped fields and calibrated probabilities make these models useful in automated extraction and classification pipelines. Applications can validate the output schema, accept high-confidence results, and send uncertain cases to a person or another system.\n\n## Five answer shapes, one endpoint\n\nEach API request supplies source material and one or more questions. The service returns a typed result for every named question, allowing a single call to extract an invoice date, locate an object, identify sensitive text, and estimate a quantity.\n\n| Task | Returned data | Example | \n|---|---|---|\n| Value extraction | Text, number, or date, with a `not found` fallback | Read the total due from an invoice | \n| Passage detection | Matching passages and their positions | Find every piece of personal information | \n| Image localization | One or more bounding boxes | Locate every damaged component | \n| Numeric estimation | An estimate and range | Count objects in a photograph | \n| Multi-select classification | Every applicable option with an individual probability | Identify all objects present in a scene | \n\nThe API accepts a recording as the material being evaluated or as the question itself. A caller can ask which language is spoken, whether a siren is audible, or submit the instruction as recorded speech instead of text.\n\n## Confidence becomes a routing signal\n\nCalibration measures how closely confidence scores match observed accuracy. Liquid reports a calibration error of 0.8 points for d1.2-ultra across 22,505 questions from 13 benchmarks, with all five compared models answering the same questions. Lower values indicate closer agreement between confidence and accuracy.\n\n| Model | Calibration error | \n|---|---|\n| d1.2-ultra | 0.8 | \n| d1.2 | 1.2 | \n| Jev 1.14 | 2.7 | \n| Perplexity Decider v1.3 | 3.7 | \n| GPT-6 Luna on the Decisions API | 7.2 | \n\nIn a well-calibrated batch, answers assigned 80% confidence should be correct about 80 times out of 100. That relationship supports operational thresholds, such as accepting results above 95% and routing the rest for review. Calibration does not establish whether the underlying accuracy is sufficient, so teams still need task-specific testing.\n\n## Internal tests cover the new ground\n\nNo public benchmark yet covers all five new answer types, according to Liquid. The company assembled held-out internal test sets from public datasets and reported the following results.\n\n| Task | Dataset | d1.2-ultra | d1.2 | \n|---|---|---|---|\n| Receipt value extraction | CORD, SROIE | 98% | 96% | \n| Personal-data detection | Nemotron PII | 94% precision, 93% recall | 92% precision, 88% recall | \n| Object localization | RefCOCO family | 94% | 94% | \n| Object count within one | COCO | 93% | 92% | \n| Multi-label object presence | COCO | 87% | 90% | \n\nAudio results ranged from 95% to 97% on ESC-50 sound identification and FLEURS language identification across 43 languages. Those evaluations used three to six answer options per question, making them narrower than open-ended audio understanding.\n\n## Pricing favors batched questions\n\nLiquid lists d1.2 at $0.02 per million input tokens and d1.2-ultra at $0.042 per million input tokens. Neither model carries an output-token charge.\n\n| Model | Input price | Output price | \n|---|---|---|\n| d1.2 | $0.02 per million tokens | $0 | \n| d1.2-ultra | $0.042 per million tokens | $0 | \n\nImage inputs consume 1.5 tokens for each 32-by-32-pixel patch at the applicable text rate. A 1024-by-1024 image therefore uses 1,536 input tokens. Multiple questions in the same request share that image input, while each question adds its own text tokens. Liquid says OpenRouter and Vercel integrations are planned.\n\n## The request is a question map\n\nThe endpoint accepts a source in the `state` field and a map of typed questions. Authentication uses a bearer token, and one request can collect several fields from the same input.\n\n``` python\nimport os\nimport requests\n\ndef extract_invoice(invoice):\n    response = requests.post(\n        \"https://api.liquid.ai/decisions/v1/systemone\",\n        headers={\n            \"Authorization\": f\"Bearer {os.environ['LIQUID_API_KEY']}\"\n        },\n        json={\n            \"model\": \"d1.2-ultra\",\n            \"state\": invoice,\n            \"questions\": {\n                \"payee\": {\n                    \"type\": \"text\",\n                    \"instructions\": \"Who issued the invoice?\"\n                },\n                \"total\": {\n                    \"type\": \"number\",\n                    \"instructions\": \"How much is due?\"\n                },\n                \"due\": {\n                    \"type\": \"date\",\n                    \"instructions\": \"By when must it be paid?\"\n                },\n                \"vat\": {\n                    \"type\": \"number\",\n                    \"instructions\": \"How much VAT is charged?\"\n                }\n            }\n        },\n        timeout=30\n    )\n    response.raise_for_status()\n    return response.json()\n```\n\n## Strong fits and exposed edges\n\nDocument processing, personal-data redaction, visual inspection, intake triage, and audio tagging align with the API's typed outputs. A single decision request may replace combinations of OCR, regular expressions, classifiers, and custom confidence heuristics. General-purpose models remain better suited to drafting, conversation, and open-ended generation.\n\nPublished benchmark figures leave several production questions unresolved, including latency, throughput, domain shift, poor scan quality, noisy audio, and calibration on a customer's own data. The d1.2 model scored 90% on multi-label object presence, three points higher than d1.2-ultra's 87%. Liquid does not explain that reversal, so exhaustive-label workloads warrant testing on both models.\n\n## Self-hosting remains a promise\n\nLiquid's models join decision-focused offerings from Perplexity, OpenAI, and Jev on the Artificial Analysis index. Their combination of constrained outputs, low input prices, and reported calibration below one point gives developers another option for extraction and classification systems that require explicit abstention and review thresholds.\n\nLiquid says it plans to release the model weights, but it has not announced a date, license, or deployment requirements. Until those details arrive, d1.2 and d1.2-ultra remain hosted API products, and any future self-hosting case depends on the terms of that release.", "url": "https://wpnews.pro/news/liquidai-s-d1-2-ultra-tops-decision-index-with-near-perfect-confidence", "canonical_source": "https://alphasignal.ai/news/liquidai-s-d1-2-ultra-tops-decision-index-with-near-perfect-confidence", "published_at": "2026-11-16 15:08:30+00:00", "updated_at": "2026-11-16 15:51:04.506083+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-products", "ai-tools", "ai-research"], "entities": ["Liquid AI", "d1.2", "d1.2-ultra", "Artificial Analysis Decision Index v2.1", "Liquid API", "Jev 1.14", "Perplexity Decider v1.3", "GPT-6 Luna"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/liquidai-s-d1-2-ultra-tops-decision-index-with-near-perfect-confidence", "markdown": "https://wpnews.pro/news/liquidai-s-d1-2-ultra-tops-decision-index-with-near-perfect-confidence.md", "text": "https://wpnews.pro/news/liquidai-s-d1-2-ultra-tops-decision-index-with-near-perfect-confidence.txt", "jsonld": "https://wpnews.pro/news/liquidai-s-d1-2-ultra-tops-decision-index-with-near-perfect-confidence.jsonld"}}