AgentsAutonomy & tool use
OpenAI case study profiles Legora's aOS legal agent platform a week after its $8.8B Series E
OpenAI's 28 October case study profiles Legora, the legal AI company valued at $8.8B, using OpenAI models, ChatGPT and Codex. Legora's aOS agent handles legal matters end to end, and its private 5,161-case benchmark suggests harness tuning can match model upgrades.
From the OpenAI News article of October 28, 2026 and cited sources: OpenAI's case study says Legora uses OpenAI models, ChatGPT and Codex to speed up legal workflows and widen contract analysis. [1] [3] [9]
SOURCES[1] OpenAI: Legora is building the agentic operating system… openai.com[3] Legal Cheek: Legora hits $8.8B valuation legalcheek.com[9] Blockchain.News: OpenAI spotlights Legora blockchain.news
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- OpenAI's case study says Legora uses OpenAI models, ChatGPT and Codex to speed up legal workflows and widen contract analysis. [1] [3] [9]
- Original source
- OpenAI News article, company source
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- October 28, 2026 article published
- Covered
- Oct 29 edition
- Evidence
- 11 sources across 7 sites
- 1OpenAI: Legora is building the agentic operating system…openai.com
- 2Legora is building the agentic operating system for legal…openai.com
- 3Legal Cheek: Legora hits $8.8B valuationlegalcheek.com
- 4Legora: Legora introduces the Legora aOSlegora.com
- 5Artificial Lawyer: Legora Launches aOSartificiallawyer.com
- 6AfterQuery: Building a frontier legal evaluation with Legoraafterquery.com
QuietHow it’s measured
Hacker News · 1 like
Reach blends public engagement (views, likes, comments, shares, stars) from the platforms where the story or its original source lives: YouTube, Hacker News, GitHub, Substack and similar. A video and the article written from its transcript share one score, taken from the video. Platforms with no data are left out rather than counted as zero, and the age of a story never lowers its score.
OpenAI's case study says Legora uses OpenAI models, ChatGPT and Codex to speed up legal workflows and widen contract analysis. The summary names no model versions and gives no performance figures. It came one week after Legora's $550M Series E, led by Atreides Management at an $8.8B post-money valuation. Legora also reports more than $200M in annual recurring revenue (ARR), a company-reported figure. [1] [3] [9]
aOS launched in May. It centres on the Legora Agent, which takes a legal matter from intake through research, drafting and review to client delivery. It runs in web, Word, Outlook and mobile, and it draws on each firm's playbooks, precedents and clause banks. Legora's legal engineers set it up for each customer, and lawyers keep control of critical decisions. [4] [5]
Legora's BAR benchmark tests the whole production system (model plus surrounding harness) rather than the bare model. It has 5,161 cases across 28 practice areas, grouped by how long the work would take a lawyer. Legal engineers write weighted rubrics, and an LLM judge scores each case over three runs. Most cases stay private to keep them out of model training data. BAR also tests Anthropic and SpaceXAI models alongside OpenAI's. [6] [10]
The evidence: about a 5% gain from harness changes versus about 3% from a new model, both on one legal benchmark. The implication for buyers and builders is that testing inside the real product, not model choice alone, may drive results.
Executive brief
OpenAI's customer story on Legora, published 28 October 2026, came out one week after Legora raised a $550M Series E at an $8.8B valuation and reported more than $200M in annual recurring revenue (ARR). The story says Legora uses OpenAI models, ChatGPT and Codex to speed up legal workflows and widen contract analysis. Its public summary does not name model versions or give numbers. The more useful technical evidence comes from Legora itself. Its "aOS" agent platform launched in May. It also built a private benchmark of 5,161 legal cases, which it uses to judge models and its own harness (the software wrapped around the model). Neither source is independent of the vendors.
What changed and event timeline
Legora launches aOS
Legora describes aOS as a connected system in which the Legora Agent handles legal matters end to end, from intake through research, drafting and review to client delivery.
BAR benchmark detailed
AfterQuery describes building Legora's Benchmark for Agentic Reasoning (BAR) with Legora. It covers 5,161 cases across 28 practice areas, and one synthetic case is public.
GPT-6 Astra tested on legal work
Legora reported about 3% overall improvement on BAR with OpenAI's new model and roughly 40% on financial-statement workflows.
Series E
Atreides Management led the $550M round at an $8.8B post-money valuation. This came about six weeks after rival Harvey raised $550M at $15.5B.
OpenAI case study
OpenAI presents Legora as a customer using its models, ChatGPT and Codex for legal workflows and contract analysis.
Capabilities and access
- Product: aOS is built around the Legora Agent.
- Access: Legora's own Legal Engineers set it up for each customer. No general-availability date has been published.
- Models: OpenAI is named as a key provider (Blockchain.News). The one exact version tied to Legora in the reviewed sources is GPT-6 Astra (Artificial Lawyer).
Read the full section
- Product: aOS is built around the Legora Agent. It runs in web, Word, Outlook and mobile, connects to tools firms already use, and draws on each firm's playbooks, precedents and clause banks (Legora).
- Access: Legora's own Legal Engineers set it up for each customer. No general-availability date has been published.
- Models: OpenAI is named as a key provider (Blockchain.News). The one exact version tied to Legora in the reviewed sources is GPT-6 Astra (Artificial Lawyer). BAR also tests Anthropic and SpaceXAI models, so Legora uses more than one model vendor (Legora BAR).
Technical analysis for researchers and developers
- What BAR measures: the whole system (model plus harness) running inside the production product, not the bare model.
- How it is graded: legal engineers write weighted rubrics covering facts, analysis, conclusions and sources.
- Reproducibility: most of the test set is private to keep it out of future model training. Only one synthetic tax case is published.
Read the full section
- What BAR measures: the whole system (model plus harness) running inside the production product, not the bare model. Cases are grouped by how long the work would take a lawyer: hours, 1–3 days, or a week or more (Legora BAR).
- How it is graded: legal engineers write weighted rubrics covering facts, analysis, conclusions and sources. An LLM judge scores each answer in a sandbox, three runs per case.
- Reproducibility: most of the test set is private to keep it out of future model training. Only one synthetic tax case is published. Outside researchers therefore cannot check most reported scores.
- For developers: Legora says harness changes alone raised output quality about 5% in one month across the same production models (AfterQuery). That suggests orchestration work matters about as much as switching models.
Claims and evidence
- Funding and adoption (company-reported, repeated by the press): $200M+ ARR, 140,000+ legal professionals, 2,400+ organizations, 80+ markets (Blockchain.News).
- GPT-6 Astra test: in Legora's demo, the agent checked financial statements across 41 documents and found all four errors that had been deliberately planted (Artificial Lawyer).
- BAR gains: these are measured on Legora's own benchmark. AfterQuery, which co-designed it, validated only the one public case (Legora BAR).
Read the full section
- Funding and adoption (company-reported, repeated by the press): $200M+ ARR, 140,000+ legal professionals, 2,400+ organizations, 80+ markets (Blockchain.News).
- GPT-6 Astra test: in Legora's demo, the agent checked financial statements across 41 documents and found all four errors that had been deliberately planted (Artificial Lawyer). This was a staged test run by a launch partner, not an independent evaluation.
- BAR gains: these are measured on Legora's own benchmark. AfterQuery, which co-designed it, validated only the one public case (Legora BAR).
- OpenAI's case study: its public summary gives no numbers on speed or contract-analysis gains (OpenAI).
Context and prior work
- History: Legora was founded in Stockholm in 2023 and was called Leya until its February 2025 rebrand (Blockchain.News).
- Valuation path: $5.55B at the March 2026 Series D, then $8.8B in October (TechCrunch).
- Harvey rivalry: Harvey has raised about $1.8B in total against Legora's roughly $1.4B (Artificial Lawyer).
Read the full section
- History: Legora was founded in Stockholm in 2023 and was called Leya until its February 2025 rebrand (Blockchain.News).
- Valuation path: $5.55B at the March 2026 Series D, then $8.8B in October (TechCrunch).
- Harvey rivalry: Harvey has raised about $1.8B in total against Legora's roughly $1.4B (Artificial Lawyer). The two even launched products both called "Intake" within hours of each other (Legal Cheek).
Limitations, safety and contested findings
- Adoption risk: Artificial Lawyer argues that aOS only pays off if firms actually trust agents to do the work, rather than trying them on occasional tasks (Artificial Lawyer).
- Human oversight: Legora's own materials keep lawyers in charge of critical decisions (Legora).
- Mixed model gains: GPT-6 Astra's overall BAR gain was about 3%, much smaller than the 40% headline on one workflow.
Read the full section
- Adoption risk: Artificial Lawyer argues that aOS only pays off if firms actually trust agents to do the work, rather than trying them on occasional tasks (Artificial Lawyer).
- Human oversight: Legora's own materials keep lawyers in charge of critical decisions (Legora).
- Mixed model gains: GPT-6 Astra's overall BAR gain was about 3%, much smaller than the 40% headline on one workflow. Artificial Lawyer called the improvement incremental but meaningful (Artificial Lawyer).
- No outside check: no independent replication of BAR results was found in the reviewed sources.
Business and practitioner implications
- Buyers: expect aOS to be rolled out with Legora's engineers, not bought as plug-and-play software.
- Builders: a domain-specific evaluation that runs inside the real product, combined with harness tuning, gives measurable gains without changing models.
- Market: Google, Anthropic and Microsoft are now entering legal AI (Artificial Lawyer).
Read the full section
- Buyers: expect aOS to be rolled out with Legora's engineers, not bought as plug-and-play software. Ask vendors for results on your own matters, not their benchmark scores.
- Builders: a domain-specific evaluation that runs inside the real product, combined with harness tuning, gives measurable gains without changing models.
- Market: Google, Anthropic and Microsoft are now entering legal AI (Artificial Lawyer). Tight ties to a model supplier, such as the OpenAI case study, may help with distribution while Legora keeps several model vendors.
The source trail.
Sources (11)
OpenAI: Legora is building the agentic operating system for legal work
openai.comLegora is building the agentic operating system for legal work
Related coverage; assess separately
openai.com