QUICK VERDICT — JULY 2026
● Cheapest available: GPT-5.6 Luna at $1/$6/M — lowest major-lab output price, 1.05M context
● Best multimodal + long context: Gemini 3.5 Flash at $1.50/$9/M — native image, video, audio, PDF, 1M context
● Best coding (until Aug 31): Claude Sonnet 5 at $2/$10/M intro — 63.2% SWE-bench Pro
● Gemini 3.6 Flash: Not launched. Registered name only. No specs or release date confirmed.
Full Comparison Table
| Model |
Status |
Input /1M |
Output /1M |
Context |
Terminal-Bench |
Best for |
| GPT-5.6 Luna |
Live |
$1.00 |
$6.00 |
1.05M |
83.2% |
High-volume routing, classification |
| Gemini 3.5 Flash |
Live |
$1.50 |
$9.00 |
1M |
76.2% |
Long docs, multimodal, Google Workspace |
| Gemini 3.6 Flash |
Not launched |
TBD |
TBD |
TBD |
Not published |
Wait for launch |
| Claude Sonnet 5 |
Live |
$2.00 intro |
$10.00 intro |
1M |
78.4% |
Coding, agents (intro ends Aug 31) |
| GPT-5.6 Terra |
Live |
$2.50 |
$15.00 |
1.05M |
87.1% |
Best mid-tier coding and reasoning |
Terminal-Bench 2.1 from Artificial Analysis July 2026. Gemini 3.6 Flash specs unconfirmed. Sonnet 5 intro pricing ends August 31, 2026.
GPT-5.6 Luna
Launched July 10, 2026. $1/$6 per million tokens — lowest output price any major lab has charged for a non-experimental model. 1.05M context. Entry tier of the GPT-5.6 family. Scored 41.3% on Nerova — handles straightforward tasks well, degrades on complex multi-step reasoning. For routing, classification, summarisation, and simple generation at high volume, Luna is the correct choice. For coding agents or complex analysis, use Terra or Gemini 3.5 Flash instead.
Gemini 3.5 Flash
$1.50/$9/M, 1M context, up to 65,000 output tokens. Scored 76.2% Terminal-Bench 2.1 and 83.6% MCP Atlas — confirmed in the official Google launch post. Accepts text, images, video, audio, and PDF natively. Strongest multimodal option in the Flash tier. Maintains reasoning quality at longer context distances where Luna degrades. Best option for processing full codebases, year-long document archives, or multi-volume research datasets in a single pass.
Gemini 3.6 Flash — What to Expect
Not available today. Google registered the model name as a potential stopgap while Gemini 3.5 Pro has now missed three consecutive launch targets. No pricing, specs, context window, or release date confirmed. Based on the naming pattern it will be an incremental Flash-tier improvement — better token efficiency and long-context reasoning, fixes to recursive tool-calling stability. Expected pricing near $1.50-$2/$9-$12/M. Not a frontier model competing with Fable 5 or GPT-5.6 Sol. When it ships, migration from 3.5 Flash is a one-line model string change.
Which to Use
High-volume routing, classification, summarisation: GPT-5.6 Luna. $1/$6/M. Avoid for complex reasoning.
Long documents, multimodal, Google Workspace: Gemini 3.5 Flash. $1.50/$9/M. Native image, video, audio, PDF. Best long-context quality in the Flash tier.
Coding and agentic tasks before August 31: Claude Sonnet 5. $2/$10/M intro. 63.2% SWE-bench Pro. Intro pricing ends August 31.
Waiting for Gemini 3.6 Flash: Stop waiting. No confirmed date. Whatever workload you are planning, build it on 3.5 Flash today.
Last updated July 18, 2026. Related: Gemini 3.6 Flash: what we know · Gemini 3.5 Pro delayed third time · GPT-5.6 Sol/Terra/Luna full review