DeepSeek V4 Flash 0731 + GPT-5.6 Luna & Terra Price Cuts

August 1, 2026

DeepSeek's official V4 Flash 0731 build is in the model picker — same 284B / 13B MoE and 1M-token context, re-post-trained for coding, reasoning, and agents at budget-tier cost. At the same time, we've cut the token rates for GPT-5.6: Luna drops from Workhorse 3× → Budget 1×, and Terra from Premium 6× → Workhorse 3×.

What you can do

• Pick DeepSeek V4 Flash from Recommended — or search for it in the picker. It takes the #3 recommended slot, replacing Gemini 3.6 Flash in the curated row (Gemini stays fully selectable). Sessions previously pinned to the older Flash slug roll forward to the 0731 build automatically.
• Run fast agentic loops cheaply — tool calling, Swarm eligibility, and Auto routing with a 1M-token context and up to 65K output tokens, on zero-retention (ZDR) endpoints.
• Use GPT-5.6 Luna at Budget 1× — the same fast high-volume Luna for chat, classification, lightweight research, and Swarm worker tasks, now billed at one-third the previous quota rate.
• Use GPT-5.6 Terra at Workhorse 3× — everyday coding and agent work at half the previous Premium rate, with the same 1.05M context and complexity-aware reasoning.
• Keep Sol unchanged — GPT-5.6 Sol stays at SOTA 12× for the hardest multi-step reasoning jobs.

Where this shows up

• You open a new chat and see the landing banner: Latest DeepSeek v4 Flash is here — one click into the model picker.
• You've been pinning Luna or Terra for daily work. Your next run costs less against plan quota — no re-pin, no migration step.
• Auto can now treat Luna as a budget-tier option, so high-volume Luna routes burn less of your allowance when the task fits.

Try it

• Pick DeepSeek V4 Flash in the picker, then: "Refactor this module across files, keeping tool use tight and the reasoning visible."
• Pick GPT-5.6 Luna, then: "Classify these files by topic, urgency, and owner, then return a concise action list."
• Pick GPT-5.6 Terra, then: "Inspect this repository, plan the refactor, implement it, and verify the risky paths."

Heads up

• Quota multipliers changed for Luna and Terra only. Luna is now Budget 1× (was Workhorse 3×); Terra is now Workhorse 3× (was Premium 6×). Sol remains SOTA 12×. Your existing plan quota and token balance do not change — each Luna/Terra token simply consumes less of it.
• DeepSeek V4 Flash stays on the Budget 1× token band.
• The old Flash slug is retired. deepseek-v4-flash aliases to deepseek-v4-flash-0731. Pinned sessions migrate automatically.
• Gemini 3.6 Flash leaves the recommended row but stays in the picker, Auto, and Swarm — nothing is removed from selection.