The Extended Brief
Sarvam Takes On Claude, Codex With Cheaper, India-Hosted Coding Agent
Brief by The AI News AI newsroom · Aug 4, 2026, 2:34 PM EDT edition
Original reporting by Inc42 — AI — Ankush Das · published Jul 30, 2026, 8:56 AM EDT
Indian engineering teams can now buy a domestically hosted coding agent that Sarvam claims solves tasks for roughly $2 each, undercutting Claude Code and Codex.
Key points
- Sarvam says Sarvam Code solved Terminal-Bench 2.1 tasks at about $2 each, versus $4.1–$27.8 for Claude Code and Codex. source ↗
- The startup also claims the agent outperforms comparable offerings from Claude and ChatGPT. source ↗
- Sarvam Code splits coding work across planner, worker, and verifier agents to better judge task completion. source ↗
- Pricing charges only for completed work, with retries and dead ends unbilled. source ↗
- The agent enters early beta on Sarvam's Indus platform, which pools credits across its agentic products. source ↗
The data
Competitor figures are the ends of the range Sarvam disclosed; the claims are self-reported.
Numbers from the original article, machine-verified against its text
From the source
“We showed that not only is it hosted in India with our own models, but actually is very competitive.”
“The same number of tasks getting solved on Sarvam Code will be cheaper.”
Retries and dead ends are not billed,” as per the startup’s disclosure.
“Sarvam Code is rolling out in early beta under the company’s Indus platform, which uses a unified credit system across its agentic AI products.”
“Pitched as an agent that builds sustainable code for the long run, Sarvam Code breaks down the coding workflow into three specialised agents — a planner, a worker and a verifier.”
Practical applications
- Run Sarvam Code's early beta on a sample of your own tickets and compare cost per solved task against your Claude Code or Codex seats.
- Independently rerun Terminal-Bench 2.1 rather than trusting Sarvam's self-reported $2 figure before committing budget.
- Use Sarvam's no-charge-for-retries policy as leverage when renegotiating per-seat coding-agent licenses.
- If India data residency drives vendor choice, evaluate the new Sarvam Inference platform for serving open-source models.
Who should care
Engineering and procurement leads at IT services firms buying coding-agent licenses, plus teams needing India-hosted AI infrastructure for data-residency reasons.
Context
Coding agents autonomously plan, write, and verify software changes, and benchmarks like Terminal-Bench score them on completed tasks. Because agents often retry, cost per solved task matters more to buyers than per-token pricing. Sarvam, a recently minted Indian AI unicorn, is pitching India-hosted models and infrastructure as its differentiator.
What to watch
- Independent replications of Sarvam's Terminal-Bench 2.1 cost and accuracy claims.
- Beta uptake by Indian IT services firms and any pricing response from Anthropic or OpenAI.
Related briefs
- Unpacking ChatGPT Work: the Agent for a Billion Users
- Agents vs. agents: how we triage HackerOne reports for $2 each, 85% as well as a human
- Announcing Cloudflare Wallets: the programmable wallet for the agentic Internet
- Popular npm Packages in the keyv and Cacheable Namespaces Compromised in Active Supply Chain Attack
Editorial score 3.4 / 5 · significance 3.0 · novelty 4.0 · edge 3.5 · perspective 3.5
Desks: Engineering · Business
Topics: tooling · business · models
Evidence basis: Reviewed from the article's full text
This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.