The price of intelligence falls while the price of control rises
GPT-5.6 Luna and Terra, Gemini Robotics ER 2 and ChatGPT for Academic Researchers, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
OpenAI
GPT-5.6 Luna and Terra
DEPLOYif you run high-volume API workloads
WHAT SHIPPEDGPT-5.6 Luna now costs 80% less and GPT-5.6 Terra 20% less, with API pricing at $0.20 per million input tokens and $1.20 per million output tokens for Luna, and $2 per million input tokens and $12 per million output tokens for Terra starting July 30.
THE CATCHLuna delivers performance comparable to models that were frontier-class a year ago at roughly 6 cents on the dollar per task, and on Agents' Last Exam it outperforms Fable 5 at an estimated cost per task nearly 99% lower.
IF YOU RUNlarge-scale document analysis, customer-interaction classification, or routine implementation at scale
SOURCEOpenAI, Advancing the price-performance frontier with GPT-5.6
Google DeepMind
Gemini Robotics ER 2
DEPLOYif you build physical AI agents
WHAT SHIPPEDGemini Robotics ER 2 is now publicly available via the Gemini API, Google AI Studio, and in private preview on Gemini Enterprise Agent Platform, with 57.4% accuracy on progress classification and 91.3% accuracy with 0.96s mean absolute distance on moment-finding.
THE CATCHIt delivers moment-finding precision at a fraction of the compute cost and 4x the execution speed of much larger model categories, and it can halt a humanoid robot when a person is nearby and resume only once the area is clear.
IF YOU RUNmulti-step robot workflows that need real-time video progress tracking
SOURCEGoogle DeepMind, Introducing Gemini Robotics ER 2
OpenAI
ChatGPT for Academic Researchers
WATCHif you work at a research institution
WHAT SHIPPEDOpenAI is giving 100,000 researchers at selected academic institutions free access to frontier models including GPT-5.6 Sol Pro, starting with 10,000 researchers this summer and expanding through 2027, with access already available at the Institute for Advanced Study and École normale supérieure.
THE CATCHGPT-5.6 Sol scores 83% on FrontierMath Tier 4 compared with 72.5% for GPT-5.5, and GPT-5.6 Sol Pro solves 31.5% of tasks on GeneBench Pro, while the program is part of a commitment of more than $250 million through 2027.
IF YOU RUNa lab that needs frontier models for genomic analysis or protein modeling
SOURCEOpenAI, Accelerating scientific discovery with ChatGPT for Academic Researchers
The Deploy Log
Tools and products
| DEPLOYif you run coding agents on Gateway DeepSeek V4 Flash. DeepSeek V4 Flash now runs on updated weights by default on AI Gateway, scoring 82.7 on Terminal-Bench, up 25.8 points from 56.9 in the April preview, with no change to the model ID or your code. Vercel |
| DEPLOYif you run small builds often Faster Vercel deployments. Vercel deployments are now up to 7 seconds faster end to end, with up to 5 seconds of fixed platform overhead removed from every build and up to 2 seconds more saved from the newest Vercel CLI. Vercel |
| WATCHif you build voice agents Grok Voice Think Fast 2.0. Grok Voice Think Fast 2.0 from xAI is now available on AI Gateway as a speech-to-speech voice model that reasons in parallel with speech and fires tool calls sooner with fewer reasoning tokens. Vercel |
| WATCHif you use ChatGPT with Vercel Sign in with ChatGPT. Sign in with ChatGPT adds your ChatGPT account as an authentication option for Vercel, available when adding the Vercel plugin to ChatGPT and when signing in to Vercel or v0. Vercel |
Left out
- Lyria 3.5: the publisher's page does not show it available yet
- Gemini in Waymo: the publisher's page does not show it available yet
- AI Gateway spend budgets: the publisher's page does not show it available yet
- MiniMax H3: the publisher's page could not be read when this edition was built
- GPT-5.6 Gateway pricing: a figure in the draft could not be found on the publisher's own page
- AI Gateway regional inference: the publisher's page does not show it available yet
- Inkling Small: it did not carry a number that moved
- Vercel Sandbox multi-user: it did not carry a number that moved
- AI Meeting Notes triggers: it did not carry a number that moved
- AI Gateway fast mode: it did not carry a number that moved
- Vercel Passport: it did not carry a number that moved
The pattern
The same week OpenAI cuts GPT-5.6 Luna to $0.20 per million input tokens and gives 100,000 researchers free access to GPT-5.6 Sol Pro, Vercel ships DeepSeek V4 Flash with a 25.8 point Terminal-Bench jump and Grok Voice Think Fast 2.0 with fewer reasoning tokens. Intelligence is becoming a cheap, ambient resource: Luna performs frontier-class work from a year ago at roughly 6 cents on the dollar, Gemini Robotics ER 2 delivers 91.3% moment-finding accuracy at 4x execution speed, and GPT-5.6 Sol scores 83% on FrontierMath Tier 4.
Control is the new premium. Vercel removes 5 seconds of fixed platform overhead and adds 2 more from its CLI, while Sign in with ChatGPT makes authentication a ChatGPT account option. The $250 million academic commitment through 2027 buys influence over who shapes the cheap models, and Gemini Robotics ER 2 halts a humanoid robot when a person is nearby. The pattern is not cheaper AI alone; it is cheaper AI plus faster, more tightly governed delivery and identity layers.