SAT, AUGUST 15, 2026
Independent · In‑Depth · Practitioner‑Tested
✎ News

AI Price War August 2026: GPT-5.6 Luna Down 80%, DeepSeek Raising Prices, Gemini Hits 1 Billion Users

August 15: OpenAI cut GPT-5.6 Luna 80% to $0.20/$1.20/M (July 30) — now ChatGPT free default with unlimited chats. Terra down 20% to $2/$12/M. Sol unchanged. DeepSeek V4 Flash raised 93% to $0.27/M. Claude Sonnet 5 rises $2→$3/M from Sept 1 (16 days). Gemini app crossed 1B monthly active users (Aug 11). Anthropic signed 20-year 191MW data centre lease at Riot Platforms Rockdale TX (~$9.1B).

By AIToolsRecap August 14, 2026 6 min read 43 views
Home Articles News Gemini AI Price War August 2026: GPT-5.6 Luna Down 80%...

AI PRICE WAR — UPDATED PRICING TABLE (AUGUST 15, 2026)

GPT-5.6 Luna: $0.20/$1.20/M (was $1/$6 — 80% cut on July 30). ChatGPT free default from Aug 6.
GPT-5.6 Terra: $2/$12/M (was $2.50/$15 — 20% cut on July 30)
GPT-5.6 Sol: $5/$30/M — unchanged
Grok 4.6: $2/$6/M — unchanged (launched Aug 12)
Claude Sonnet 5: $2/$10/M → $3/$15/M from Sept 1 (16 days)
DeepSeek V4 Flash: $0.14→$0.27/M input (93% increase — effective Aug 14)
Gemini 3.6 Flash: $1.50/$7.50/M — unchanged
Claude Opus 5: $5/$25/M — unchanged
GPT-5.6 Sol Fast: $10/$60/M — NEW (2.5x speed of standard Sol at 2x price)
Gemini app: 1 billion monthly active users (announced Aug 11 by Sundar Pichai)

The OpenAI Price Cuts — What Changed and Why

Per VentureBeat's analysis of the cuts, OpenAI reduced GPT-5.6 Luna by 80% on July 30, 2026 — from $1/$6/M to $0.20/$1.20/M — attributing the reduction to efficiency improvements from internal model development, including the model's role in optimizing production software and improving speculative decoding. Terra fell 20% from $2.50/$15/M to $2/$12/M. Sol's pricing remains unchanged at $5/$30/M. As Weckr's pricing analysis notes, the Luna cut has a subtle but significant implication: Luna at $0.20/$1.20/M now undercuts gpt-5.4-nano ($0.20/$1.25/M) on output price while being a more capable model. "The floor of the OpenAI lineup effectively moved up a capability tier at the same price." Any team that routed traffic to gpt-5.4-mini ($0.75/$4.50/M) purely for cost reasons now has a better option at the same capability tier.

Per Tech Insider Canada's full timeline, the cuts coincided with OpenAI making Luna the default model for free and Go-tier ChatGPT users starting August 6, with unlimited text chats and a "Think" reasoning toggle. The timing maps directly onto OpenAI's competitive pressure: TechSpot reports that ChatGPT's global web traffic share fell from ~79% in May 2025 to 53.9% in May 2026, with Gemini climbing to 27.7% of AI assistant app users. Making a capable model free and unlimited is OpenAI's retention lever.

DeepSeek Is Raising Prices While Everyone Else Cuts

The move that inverts the expected pattern: DeepSeek raised deepseek-v4-flash input prices from $0.14/M to $0.27/M effective August 14 — a 93% increase. Per Tech Startups's August 14 roundup, "OpenAI and Anthropic are lowering prices on some models as lower-cost Chinese competitors including DeepSeek and Moonshot AI gain users among companies looking to control increasingly large inference bills" — and simultaneously DeepSeek is raising its prices as demand exceeds its infrastructure capacity. The deepseek-v4-flash that was the cheapest capable agentic model at $0.14/M on July 31 is now $0.27/M. Still cheaper than most alternatives, but the gap has narrowed significantly. Combined with the October 24 deprecation deadline for deepseek-chat and deepseek-reasoner, this week's price increase adds urgency to the migration calculus.

Gemini Crosses 1 Billion Monthly Active Users

Per AI Weekly's coverage, Sundar Pichai announced on August 11 that the Gemini app has crossed 1 billion monthly active users — Google's fastest-growing product ever and its 14th product to hit the billion-user mark alongside Search, Gmail, Android, and YouTube. TechCrunch reports 63% of users engage with voice features. The milestone arrives as OpenAI reports its own 1 billion monthly active users (crossed June 2026). For the first time in AI's consumer history, two AI assistants are simultaneously at the 1 billion MAU mark.

Updated Routing Guide — What to Use at Each Price Point Now

Use caseBest model nowInput priceNotes
High-volume classification/routingGPT-5.6 Luna$0.20/MWas $1/M before July 30. Now OpenAI's cheapest.
Mid-tier general tasksGemini 3.6 Flash$1.50/M1M context, Search grounding, 304 t/s
Writing/long-form (act before Aug 31)Claude Sonnet 5$2/M now → $3/M Sept 1Best writing quality. 16 days at $2/M.
Agentic coding under 200K tokensGrok 4.6$2/MAPEX-Agents leader. Doubles above 200K.
Frontier reasoning / hard tasksClaude Opus 5$5/MAA Index 61, ARC-AGI-3 30.2%, stable pricing
Repo-scale coding + async PRGPT-5.6 Sol$5/MDeepSWE leader, Codex async PR

Anthropic's 20-Year Riot Platforms Data Centre Lease

Alongside the Theseus Infrastructure JV announced August 10, Riot Platforms disclosed a 20-year data centre lease at its Rockdale, Texas campus — Bloomberg identified the tenant as Anthropic. The deal covers 191 MW, runs through June 2048, and carries approximately $9.1 billion in base revenue with two five-year extension options that could push total value to $16.1 billion. This is separate from Theseus (which targets new purpose-built facilities) and the existing AWS $100B+ commitment. Anthropic is simultaneously building its infrastructure stack through multiple channels: Macquarie/GIC JV for new construction, long-term leases for existing capacity (Riot/Rockdale), and cloud commitments to AWS for on-demand access.

Sources: VentureBeat price war analysis · Weckr routing guide · TechSpot — 1B users + price cuts · Quartz — OpenAI 1B announcement · Tech Insider Canada — Luna free default · AI Weekly — Gemini 1B users · Tech Startups August 14 roundup · Related: Sonnet 5 Aug 31 deadline → · Anthropic Theseus data centres →

Tags
AI NewsGenerative AIChatGPTGeminiGoogle2026

Spot an inaccuracy?

We verify facts before publishing and correct errors promptly. If something in this article is wrong or outdated, let us know.

Report an error →