# AI21 Labs presents Jurassic-X, a system combining language models with external tools [Canonical reading](https://escapemeridian.com/readings/ai21-labs-presents-jurassic-x-a-system-combining-language-b1ca7e37) AI21 Labs published its presentation of Jurassic-X on April 9, 2022. The company describes Jurassic-X as its implementation of a Modular Reasoning, Knowledge and Language (MRKL) system, which combines language models with external knowledge sources and symbolic reasoning experts. It says a router can direct parts of a request to relevant tools, such as databases, Wikidata or a calculator, and that the system can show intermediate steps. AI21 Labs presents this design as a way to handle tasks it says language models struggle with, including accessing current information, performing calculations and answering questions that require several steps. The company describes examples involving database queries, current affairs, math, weather and currency exchange rates. It also says the system can display which expert handled each part of a request. These are the company’s descriptions and demonstrations, not evidence here of performance across real-world uses. The post notes challenges including training discrete experts, smoothing their interface with the neural network and routing among modules. How reliably does Jurassic-X select the right expert and produce accurate answers when connected to changing or private data? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source AI21 Labs · Article Original source date: Apr 9, 2022 (2022-04-09T06:06:37Z). Date basis: Date from the publisher’s own records, through its API. Added to Escape Meridian: 2026-10-29T16:18:35.459Z. First observed: 2026-10-27T15:47:54.052Z. Retrieved: 2026-10-27T15:49:41.130Z. - [Jurassic-X: Crossing the neuro-symbolic chasm with the MRKL system](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Limits Text as retrieved on October 27, 2026; originally published April 9, 2022. The post presents the capabilities and examples as AI21 Labs’ descriptions; the supplied evidence does not establish their performance across real-world uses. The retrieved page is marked as edited after publication, so later additions may be present; the evidence does not identify which passages were added. ## Open question How reliably does Jurassic-X select the right expert and produce accurate answers when connected to changing or private data? ## Source-stated AI21 Labs describes Jurassic-X as its implementation of a MRKL system that augments language models with external knowledge sources and symbolic reasoning experts. > MRKL systems include one or more language models, and augment them with external knowledge sources as well as symbolic reasoning experts that can handle tasks that lie beyond the reach of neural models. - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source-stated AI21 Labs says Jurassic-X routes parts of a request to specialized resources, such as a wiki API, calendar, database and calculator, before formatting an answer with a language model. > the list of “Green energy companies” is routed to Wiki API, “last month” dates are extracted from the calendar and “share prices” from the database. The “largest increase“ is computed by the calculator - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source-stated AI21 Labs presents database access, current information, math and multi-step questions as tasks Jurassic-X can address through connections to external resources and experts. > Jurassic-X can decompose such problems into the basic questions, route each to the relevant expert, and put together an answer in free language. - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source-stated AI21 Labs says the system displays which expert answered each part and the intermediate steps taken. > Jurassic-X is designed to be more transparent by displaying which expert answered which part of the question, and by presenting the intermediate steps taken - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source-stated AI21 Labs identifies technical challenges in training experts, integrating them with the neural network and routing among modules. > training the discrete experts, smoothing the interface between them and the neural network, routing among the different modules - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source and contributor context AI21 Labs · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Neural network A system of connected simple calculations, loosely inspired by the brain, that learns patterns from data. - [Definition reference](https://developers.google.com/machine-learning/glossary#neural-network) ## AI21 AI21 publishes technical research and product announcements about its language models and AI systems. - [Explore this context](/entities/ai21) - [Reference](https://www.ai21.com/blog/) ## Editorial provenance Version 1. Publication ID: dd0a7c90-7935-32cc-dc8d-32ab81924a6f. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T16:17:56.738Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 647243735886957f0fed4a249d2cd6a18fbde7ffefe140828f6d760d5a8e564c. Source text hash: 5a559b7fa0fad454e397ef537e7a524b241fa053a7932bc6a49b24b43dbabc6b. ## Revision and source-check history ### Version 1 — 2026-10-29T16:18:35.459Z AI21 Labs published its presentation of Jurassic-X on April 9, 2022. The company describes Jurassic-X as its implementation of a Modular Reasoning, Knowledge and Language (MRKL) system, which combines language models with external knowledge sources and symbolic reasoning experts. It says a router can direct parts of a request to relevant tools, such as databases, Wikidata or a calculator, and that the system can show intermediate steps. AI21 Labs presents this design as a way to handle tasks it says language models struggle with, including accessing current information, performing calculations and answering questions that require several steps. The company describes examples involving database queries, current affairs, math, weather and currency exchange rates. It also says the system can display which expert handled each part of a request. These are the company’s descriptions and demonstrations, not evidence here of performance across real-world uses. The post notes challenges including training discrete experts, smoothing their interface with the neural network and routing among modules. How reliably does Jurassic-X select the right expert and produce accurate answers when connected to changing or private data? ## Original source AI21 Labs · Article Original source date: Apr 9, 2022 (2022-04-09T06:06:37Z). Date basis: Date from the publisher’s own records, through its API. Added to Escape Meridian: 2026-10-29T16:18:35.459Z. First observed: 2026-10-27T15:47:54.052Z. Retrieved: 2026-10-27T15:49:41.130Z. - [Jurassic-X: Crossing the neuro-symbolic chasm with the MRKL system](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Limits Text as retrieved on October 27, 2026; originally published April 9, 2022. The post presents the capabilities and examples as AI21 Labs’ descriptions; the supplied evidence does not establish their performance across real-world uses. The retrieved page is marked as edited after publication, so later additions may be present; the evidence does not identify which passages were added. ## Open question How reliably does Jurassic-X select the right expert and produce accurate answers when connected to changing or private data? ## Source-stated AI21 Labs describes Jurassic-X as its implementation of a MRKL system that augments language models with external knowledge sources and symbolic reasoning experts. > MRKL systems include one or more language models, and augment them with external knowledge sources as well as symbolic reasoning experts that can handle tasks that lie beyond the reach of neural models. - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source-stated AI21 Labs says Jurassic-X routes parts of a request to specialized resources, such as a wiki API, calendar, database and calculator, before formatting an answer with a language model. > the list of “Green energy companies” is routed to Wiki API, “last month” dates are extracted from the calendar and “share prices” from the database. The “largest increase“ is computed by the calculator - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source-stated AI21 Labs presents database access, current information, math and multi-step questions as tasks Jurassic-X can address through connections to external resources and experts. > Jurassic-X can decompose such problems into the basic questions, route each to the relevant expert, and put together an answer in free language. - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source-stated AI21 Labs says the system displays which expert answered each part and the intermediate steps taken. > Jurassic-X is designed to be more transparent by displaying which expert answered which part of the question, and by presenting the intermediate steps taken - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source-stated AI21 Labs identifies technical challenges in training experts, integrating them with the neural network and routing among modules. > training the discrete experts, smoothing the interface between them and the neural network, routing among the different modules - [Supporting source](https://www.ai21.com/blog/jurassic-x-crossing-the-neuro-symbolic-chasm-with-the-mrkl-system/) ## Source and contributor context AI21 Labs · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Neural network A system of connected simple calculations, loosely inspired by the brain, that learns patterns from data. - [Definition reference](https://developers.google.com/machine-learning/glossary#neural-network) ## AI21 AI21 publishes technical research and product announcements about its language models and AI systems. - [Explore this context](/entities/ai21) - [Reference](https://www.ai21.com/blog/) ## Editorial provenance Version 1. Publication ID: dd0a7c90-7935-32cc-dc8d-32ab81924a6f. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T16:17:56.738Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 647243735886957f0fed4a249d2cd6a18fbde7ffefe140828f6d760d5a8e564c. Source text hash: 5a559b7fa0fad454e397ef537e7a524b241fa053a7932bc6a49b24b43dbabc6b. ### passed — 2026-10-29T16:17:56.738Z — anthropic/claude-opus-5.5 Publication: dd0a7c90-7935-32cc-dc8d-32ab81924a6f attribution — pass: The reading consistently attributes the MRKL description, capability claims and examples to AI21 Labs. This matches the publisher page, which presents Jurassic-X as the company's own implementation. > Jurassic-X, AI21 Labs’ implementation of it summary — pass: The title, summary and question accurately reflect the post. The reading names the actor, expands and explains MRKL, and describes the router and tools in plain terms. It conveys the claimed significance (current data, math, multi-step questions, transparency) as company claims. The question follows from the routing challenges the post itself names and is not leading. > routing among the different modules date — pass: originalPublished is 2022-04-09T06:06:37Z, so the stated date of April 9, 2022 matches the UTC day. Nothing is presented with hindsight or as happening after that date. The cited body text is internally consistent with an early-2022 original, citing Jurassic-1's August 2021 release and the presidency example. The page is marked as edited, and the reading discloses this in its limitations. The evidence is a publisher page, so no archive disclosure is needed. > In August 2021 we released Jurassic-1, a 178B-parameter autoregressive language model. support — pass: Every claim is backed by verbatim source text and framed as AI21's description. The reading explicitly declines to treat the demonstrations as evidence of real-world performance. The cited passages are core article body consistent with the 2022 original rather than evident later additions, and the reading flags the post-publication edit. No benefits or costs are invented. > Jurassic-X can decompose such problems into the basic questions, route each to the relevant expert, and put together an answer in free language. media — pass: No media was selected, and no verified media candidates were supplied, so selecting none is appropriate. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/b1ca7e374c6c3ad90e7bf8ae86c46a76eaae971162934f7dd3be0dddf0f80734/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/ai21-labs-presents-jurassic-x-a-system-combining-language-b1ca7e37.json)