AI Daily Brief: GPT-5.6 Steps Into the Open, Google Rebuilds Gemini From Scratch

For once the interesting news is not a lawsuit or a funding round. The frontier is moving again: OpenAI's newest family is now open to everyone, Google has torn its next model down to the studs, and the chip industry spent the week pairing off into alliances that would have been unthinkable a year ago.

GPT-5.6 steps into the open

OpenAI's GPT-5.6 family, Sol at the flagship end, Terra for everyday work and Luna for speed and cost, is now available to the general public after an unusual staggered rollout that began behind a government review. The most talked-about capability is also the most uncomfortable one: the models are strong enough at finding software vulnerabilities that the release process itself was shaped around that risk.

The improvement is real and it cuts both ways, since the same skill that worries security agencies also gives defenders their best patching tool yet. The question enterprises will answer over the next quarter is whether a vulnerability-hunting model gets treated as a security product or a security problem.

Google chose to rebuild Gemini rather than ship it

Gemini 3.5 Pro is targeting July 17 for general availability, and the story behind the date is more revealing than the date itself. Google reportedly scrapped the Gemini 2.5 Pro base model entirely and ran a fresh pre-training cycle, concluding the old architecture could not close its gaps in mathematical reasoning and image quality. The specs in circulation, a 2-million-token context window and a Deep Think reasoning layer among them, all come from leaks; as of Monday there was no model card, no pricing page and no API listing.

A company that would rather restart pre-training than ship a compromised flagship is telling you something about how it thinks this race is scored. Thursday will show whether the rebuild bought Google a lead or just time.

Nvidia's new hedge is partnering with its rivals

The Information reports that Nvidia is combining its hardware with chip startup d-Matrix in a joint system for running AI models, the clearest sign yet of a strategic turn from defending every layer to organising the ecosystem around itself. It follows a new programme in which Nvidia taps its cloud partners to supply startups with compute in exchange for revenue-sharing, and a multiyear pact with SK hynix on memory for its Vera Rubin systems.

Read together with Meta's Broadcom-designed Iris chip entering production in September, the pattern is a chip market reorganising into alliances. Nvidia has apparently concluded that if custom silicon and specialised inference chips are inevitable, it would rather have equity in the arrangement than a competitor.

Cerebras takes the inference fight to Europe

Cerebras will spend several billion dollars building European data centres, starting in France and the Nordics, targeting 200MW of capacity by the end of 2027. Part of that capacity is earmarked for OpenAI workloads under their $20 billion-plus compute agreement, but the stated driver is sovereign demand: European governments and enterprises want low-latency AI capacity that does not sit in the US or Asia.

Sovereign compute has quietly become the growth story in AI infrastructure, and it rewards exactly the kind of specialised, deployable capacity that wafer-scale players sell.

What it means for India

Every thread above lands somewhere in India. Model convergence at the frontier, with GPT-5.6 public and Gemini rebuilt, keeps pushing the differentiation down into deployment, which is the layer Indian IT sells. The earnings tape agrees: HCLTech followed TCS by posting a 20.3% jump in first-quarter profit on revenue up 13.9%, the second Indian IT major in two days to beat estimates on AI-heavy demand. And the sovereign-compute wave Cerebras is riding in Europe is the same argument behind IndiaAI's domestic GPU build-out: inference capacity is becoming national infrastructure, and countries that host it capture the workloads. The alliance turn in chips also lowers the barrier for India's design and packaging ecosystem, since a market organised around partnerships has more seats at the table than one organised around a single vendor.

Markets and AI money

Tuesday opened rough: Gift Nifty signalled a gap-down start near 24,050 as crude spiked to a one-month high, putting Monday's IT-led recovery straight back under pressure. The sector carries the burden of proof today: HCLTech's numbers extend the AI-demand story TCS started, and the market gets to decide whether two beats make a trend.

Metric	Level	Move
Gift Nifty (pre-open)	24,050	-0.69%
Sensex (Mon close)	77,616.40	+0.06%
Nifty IT (Mon)	-	+3.59%
HCLTech Q1 net profit	Rs 4,624 cr	+20.3% YoY
HCLTech Q1 revenue	Rs 34,579 cr	+13.9% YoY

The models are shipping, the chipmakers are choosing partners, and the money is watching Dalal Street's IT screens today. That's the brief for Tuesday.