BusinessMarkets, money & strategy
Atlassian makes a spending commitment to OpenAI, but GPT-6 Astra will not be Rovo's default model
Atlassian has committed to spending on OpenAI models and added ChatGPT and Codex plugins for Jira, Confluence and Bitbucket. GPT-6 Astra will not be Rovo's default model, though. Atlassian's assistant keeps sending each task to different providers, including Anthropic.
OpenAI called the expanded deal effectively a spending commitment and says its models will power agents across Rovo and Atlassian's platform. Rovo stays multi-model, however. An internal gateway picks a provider for each task based on capability, speed and cost. Atlassian's AI transparency page lists GPT, Claude, Gemini and Llama models behind Rovo Chat. [1] [2] [8]
Available now: plugins that give ChatGPT and Codex access to Jira, Confluence and Bitbucket context through the Teamwork Graph, plus a rebuilt MCP server. Atlassian says the graph holds more than 150 billion connections. The MCP server and command-line access have been in open beta since May, and Jira Service Management support is listed as coming soon. [1] [2] [5] [6]
The headline feature, agents that pick up Jira work items on their own, has no launch date or price. The size and financial terms of the spending commitment are also undisclosed. [3]
Astra-backed workflows need admin work first. Astra is switched off by default in enterprise workspaces and costs $10 per million input tokens and $50 per million output tokens through the API. OpenAI reportedly rated it Critical under its Preparedness Framework. Reporting also points to prompt-injection risk, misalignment monitoring that can interrupt legitimate work, and unclear criteria for Zero Data Retention eligibility. [7] [2]
Rovo routes tasks by capability, speed and cost, and its usage figures are vendor-reported. Implication: enterprises get OpenAI access without lock-in, while Atlassian presents its Teamwork Graph, not any single model, as the lasting asset.
Atlassian and OpenAI expand partnership: research dossier
Executive brief
Atlassian has agreed to spend money on OpenAI's models, which OpenAI described as "effectively a spend commitment". Even so, OpenAI's newest flagship, GPT-6 Astra, will not be the default model in Atlassian's AI assistant, Rovo. Rovo keeps sending each task to whichever provider fits it best, and that includes Anthropic (VentureBeat). The deal was announced on October 6, 2026. It gives Atlassian wider access to the GPT-6 and GPT-5.6 models and adds plugins that let ChatGPT and Codex read Jira, Confluence and Bitbucket. The headline idea, AI agents picking up Jira tickets on their own, has no date or price yet (Cellcog).
What changed and event timeline
Partnership begins
Atlassian starts working with OpenAI. The 2026 expansion builds directly on that relationship, according to coverage of the announcement ().
Teamwork Graph opened to outside agents
At Team '26, Atlassian opened its Teamwork Graph (its map of people, work and documents) through an MCP server and a command-line tool. Both are in open beta, and Codex and Claude Code are named as supported clients (;).
GPT-6 Astra launches
OpenAI released Astra and then widened access. Enterprise customers got it right away, but it ships switched off by default for each workspace ().
Expanded partnership announced
Under the new deal, OpenAI frontier models power agents across Atlassian's platform and Rovo. New plugins connect ChatGPT and Codex to Atlassian's tools (;).
Capabilities and access
- Models: GPT-6 Astra and the GPT-5.6 series (Cellcog).
- Available now:
- OpenAI models help power Rovo's reasoning.
Read the full section
- Models: GPT-6 Astra and the GPT-5.6 series (Cellcog).
- Available now:
- OpenAI models help power Rovo's reasoning.
- ChatGPT and Codex can reach Jira, Confluence and Bitbucket context through plugins backed by the Teamwork Graph.
- A rebuilt MCP server is live.
- Support for Jira Service Management is "coming soon" (VentureBeat).
- Planned: agents picking up Jira work items. No launch date or price has been given (Cellcog).
- Astra API pricing: $10 per million input tokens and $50 per million output tokens (RuntimeWire).
Technical analysis for researchers and developers
- Routing: Rovo's internal AI gateway chooses a provider for each task based on capability, speed and cost.
- Context layer: Atlassian says the graph holds more than 150 billion connections.
- Astra specifications: a context window of about 1.05 million tokens, up to 128,000 output tokens, and tools for web search, code execution and computer use (RuntimeWire).
Read the full section
- Routing: Rovo's internal AI gateway chooses a provider for each task based on capability, speed and cost. Models are treated as swappable parts, and the Teamwork Graph is the piece meant to last (VentureBeat).
- Context layer: Atlassian says the graph holds more than 150 billion connections. Access runs through the MCP server or the command-line tool, both in open beta (IT Brief UK).
- Astra specifications: a context window of about 1.05 million tokens, up to 128,000 output tokens, and tools for web search, code execution and computer use (RuntimeWire).
- Evaluation: no task benchmarks for the integration have been published.
Claims and evidence
No independent evaluation of how well the integration performs is available.
Read the full section
| Claim | Status |
| More than 3,000 Atlassian developers use Codex (Cellcog) | Vendor-reported. Both companies give the same figure, but that is not independent confirmation. |
| Rovo is used by more than 75% of the Fortune 500 and 90% of Atlassian's enterprise cloud customers (IT Brief UK) | Vendor-reported |
| 14 million+ Rovo-assisted actions in one month; agent automations up sevenfold in six months (IT Brief UK) | Vendor-reported |
| Astra is not Rovo's default model; Rovo stays multi-model (VentureBeat) | Vendor statement, carried in independent reporting |
No independent evaluation of how well the integration performs is available.
Context and prior work
- Multi-model from the start: Atlassian's AI transparency page lists GPT, Claude, Gemini and open-source Llama models behind Rovo Chat (Atlassian).
- Rovo's origin: Atlassian launched Rovo at Team '24 (Constellation Research).
- Anthropic ties: Anthropic's Scott White called Atlassian's MCP integration one of the most-used enterprise MCP integrations on Claude (VentureBeat).
Limitations, safety and contested findings
- Safety rating: OpenAI reportedly rated Astra "Critical" under its Preparedness Framework, because it can find previously unknown software vulnerabilities and develop ways to exploit them (RuntimeWire).
- Monitoring side effects: OpenAI's misalignment monitoring can slow, pause or stop legitimate work (VentureBeat).
- Other reported risks: prompt-injection risk remains, Astra's reasoning is harder to audit than earlier models, and the criteria for Zero Data Retention (no stored data) eligibility are unclear (VentureBeat).
Read the full section
- Safety rating: OpenAI reportedly rated Astra "Critical" under its Preparedness Framework, because it can find previously unknown software vulnerabilities and develop ways to exploit them (RuntimeWire).
- Monitoring side effects: OpenAI's misalignment monitoring can slow, pause or stop legitimate work (VentureBeat).
- Other reported risks: prompt-injection risk remains, Astra's reasoning is harder to audit than earlier models, and the criteria for Zero Data Retention (no stored data) eligibility are unclear (VentureBeat).
- Undisclosed terms: the dollar amount and financial terms have not been published (Cellcog).
Business and practitioner implications
- No lock-in: customers are not tied to OpenAI. Rovo's routing keeps choosing between providers, so the Teamwork Graph is the strategic asset being sold.
- Admin work first: before using Astra-backed workflows, admins must turn on Astra access (RuntimeWire).
- What developers can do today: connect Codex or ChatGPT to Jira and Confluence context through the plugins and the MCP server.
Read the full section
- No lock-in: customers are not tied to OpenAI. Rovo's routing keeps choosing between providers, so the Teamwork Graph is the strategic asset being sold.
- Admin work first: before using Astra-backed workflows, admins must turn on Astra access (RuntimeWire). They should also check data-retention terms and budget for the high per-token price.
- What developers can do today: connect Codex or ChatGPT to Jira and Confluence context through the plugins and the MCP server.
- What to wait for: don't plan around autonomous ticket pickup until a date and price are announced.
Sources
Read the full section
- OpenAI: Atlassian and OpenAI expand partnership
- VentureBeat: Atlassian deepens its OpenAI partnership with a spend commitment
- Crypto Briefing: Atlassian and OpenAI expand partnership
- Cellcog: Atlassian OpenAI Partnership
- IT Brief UK: Atlassian opens Teamwork Graph to third-party AI tools
- Nasdaq: Atlassian Opens Up Its Teamwork Graph
- RuntimeWire: OpenAI expands GPT-6 Astra
- Atlassian AI transparency
- Constellation Research: Atlassian launches Rovo