Sarvam AI marketed its new coding agent as a cheaper, India-built rival to Claude Code and OpenAI's Codex. A technical teardown found it's actually OpenAI's Codex, wearing a new name.
Sarvam AI launched Sarvam Code on July 30 at Epoch, its first developer conference, in Bengaluru. The pitch was simple: a three-agent system, a planner, a worker and a verifier, that Sarvam said solved 72 of 89 Terminal-Bench 2.1 tasks for roughly $2 each, undercutting Claude Code's $4.1 and OpenAI Codex's $27.8 per task, according to Inc42's coverage of the launch. That same day, a teardown published at ojasviyadav.com pulled apart the 348-megabyte binary Sarvam shipped and found something the launch materials never mentioned: OpenAI's own open-source Codex CLI, repackaged under a new name.
The overlap isn't a vague resemblance. It's line-for-line.
The analysis, done through static tools like otool, nm and strings rather than by running the software, found Sarvam Code sharing Codex's exact command surface: exec, resume, fork, mcp. It found the same three sandbox modes, read-only, workspace-write and danger-full-access, and the same approval and shell_environment_policy structure in its config schema. Buried in the binary was an un-scrubbed HTTP header string, x-openai-subagent, a piece of OpenAI's own internal plumbing that should never have survived a rewrite. Even Codex's most dangerous flag, --dangerously-bypass-approvals-and-sandbox, persists verbatim.
What Sarvam actually built, according to the teardown, sits on top of that base: OCR and document digitization tuned for 22 scheduled Indian languages, a Guardian approval layer, something called RLM mode, and a self-improvement governor. Real engineering, in other words. Just not the part being sold as the breakthrough.
Sarvam AI is India's newest AI unicorn after raising $234 million from HCLTech Bengaluru-based Sarvam AI has raised $234 million in the first close of its Series B round led by HCLTech, which wrote a $150 million check for a 10.46% stake and pushed the startup's valuation to $1.5 billion. The deal pairs HCLTech's enterprise relationships across banking, insurance, and government with Sarvam's India-trained language models... - Indian AI startup raises $234 million from HCLTech - Sarvam AI becomes India's newest AI unicorn valuation
This finding traces to one independent teardown, not to OpenAI or a security firm, and it hasn't been picked up by major outlets. Neither Sarvam AI nor OpenAI has publicly addressed it. That's worth weighing before treating this as a settled scandal. But it doesn't change what the teardown documents inside the binary itself: identical strings, identical schemas, identical flags, none of them credited.
The license actually allows this. Attribution is the catch. #
OpenAI's Codex CLI sits on GitHub under an Apache 2.0 license, with more than 400 contributors and roughly 75,000 stars. Apache 2.0 is generous. It lets anyone fork the code, rebrand it, sell it, even keep their own changes closed. What it doesn't let you do is strip the original copyright notice and license text out of the distribution, or pass the work off without saying what you changed. If the teardown is right that Sarvam shipped no attribution to OpenAI or the Codex project anywhere in Sarvam Code, that isn't just an awkward look. It's the one thing the license actually requires.
Sarvam AI isn't a fringe player playing fast and loose to save money. It's Bengaluru's best-funded AI bet, founded in August 2023 by Vivek Raghavan and Pratyush Kumar. It closed the first $234 million of a $300 million Series B in June, at a $1.5 billion valuation, with HCLTech putting in $150 million and Bessemer Venture Partners joining longtime backers Khosla Ventures and Peak XV Partners. India's government picked Sarvam, out of 67 applicants, to build the country's sovereign foundational LLM, handing it access to 4,096 Nvidia H100 GPUs. This is a company with the money, and the government mandate, to build its own coding agent from scratch.
It chose not to, at least not entirely. Cofounder Pratyush Kumar framed Sarvam Code at launch as proof that Indian IT firms no longer need to pay Anthropic for coding tools. "IT services companies are buying licenses from Anthropic and so on because they want to do coding," he said, according to Inc42. "We showed that not only is it hosted in India with our own models, but actually is very competitive." That pitch reads very differently once you know the tool doing the competing was, underneath its new name, built by the company Sarvam was supposedly beating on price.
Sovereign AI is the framing every government-adjacent AI startup reaches for now: build it here, host it here, don't depend on Silicon Valley. Sarvam Code shows how easily that framing slips when the fastest path to a competitive product is still someone else's code, wearing a local coat of paint.
Also read: AI Watched a Live Brain Operation in Real Time and Helped Save a Man's Sight • Google DeepMind Is Losing the AI Talent War It Used to Win Easily • Salesforce Stock Jumps 14% as Benioff's AI Numbers Silence SaaS Doubters