← All digests

2023-03-14

India AI Digest — Tuesday, March 14, 2023

  • OpenAI ships GPT-4. Multimodal input, step-change reasoning, and pricing roughly 10× GPT-3.5 Turbo per token. Khan Academy launches Khanmigo as the named partner.
  • Position: foundation_model_capability -1 (3) — frontier resets; gap to Indian foundation-model effort widens substantially. consumer_adoption_depth +1 (2) — ChatGPT Plus seeds the Indian paying-consumer wedge.

Archive entry. Written retrospectively in April 2026 as part of historical backfill. The contemporaneous voice is preserved as if filed on the event date.

OpenAI ships GPT-4; Khan Academy unveils Khanmigo as launch partner

OpenAI released GPT-4 on March 14, 2023. The model is multimodal, accepting image and text inputs and producing text outputs, with reported step-change improvements over GPT-3.5 on reasoning, coding, and domain-specific benchmarks — Bar exam, MMLU, multiple Olympiad-level tasks. Headline API pricing is approximately $0.03 per 1K input tokens and $0.06 per 1K output tokens [TBV against the launch-day pricing page] — meaningfully higher per-token than GPT-3.5 Turbo. Khan Academy is named as a launch partner with the Khanmigo tutor demo.

What this means. The capability jump is real. The pricing reframes who can build what.

GPT-4's reasoning gain materially changes outcomes for legal contract review, code generation, document understanding, and complex enterprise support. The higher cost is justified there. For workloads where GPT-3.5 was good enough — most consumer chat, content generation, basic Q&A — GPT-4 is a capability ceiling that most builders will not pay for at scale.

The harder question for the Indian stack is not whether GPT-4 is impressive. It is whether the new pricing tier is reachable on Indian unit economics. The answer is sector-specific.

India angle. Four sectors are reading GPT-4 in opposing ways this week.

  • SI layer (TCS, Infosys, Wipro, HCL). This is where India's AI economic exposure concentrates and where the capability jump matters most. GPT-4 rebundles what enterprise AI services means in client conversations. The offerings catalog has to be rewritten. Existing GPT-3.5-based proofs of concept become benchmarks to beat, not deliverables. Net: enabling, but it forces fast catalog churn at SI scale.
  • Indian conversational AI vendors (Haptik, Yellow.ai, Observe.AI, Uniphore). Strategic threat alongside opportunity. Global enterprise clients who had been buying conversational-AI offerings can now consider building in-house against the OpenAI API. Vertical tuning and Indian-context expertise are still real; the moat compresses.
  • Indian edtech (Byju's, Unacademy, Vedantu, smaller AI-tutoring plays). The Khanmigo signal — Khan Academy launching an AI tutor specifically on GPT-4 — is directional. The unit-economics math is unforgiving. At the headline pricing, a 5K–10K-token tutoring session costs roughly $0.15–$0.50 in raw inference. The Indian consumer edtech ARPU ceiling sits at ₹80–200 monthly — about $1–$2.50 at March 2023 rates. Even modest session frequency exceeds the entire revenue per user. The path forward is institutional / school-district sales, aggressive caching with GPT-3.5 fallback, or waiting for materially cheaper inference.
  • BFSI. Capability is attractive; deployability is constrained. RBI-regulated workloads cannot move to GPT-4 without resolving data residency. As of March 2023, OpenAI does not offer India-region inference. Indian banks and NBFCs that want GPT-4-class capability in customer-facing or regulated contexts have to wait for a residency answer or stay on GPT-3.5-based offerings deployed via approved cloud architectures.

Source: OpenAI GPT-4 release, March 14, 2023; OpenAI API pricing page as of March 2023 [link TBV]; Khan Academy Khanmigo launch announcement [link TBV]. → Technical Report

Confidence: medium — release date, multimodal input, and headline benchmark categories are well-attested; specific pricing values pending re-verification against the source page.