PRICING SNAPSHOT 2026-09-06CLAUDE FABLE 5 $10.00/$50.00 MTOKCLAUDE FABLE 5.1 $10.00/$50.00 MTOKCLAUDE OPUS 5 $5.00/$25.00 MTOKCLAUDE OPUS 4.8 $5.00/$25.00 MTOKCLAUDE SONNET 5 $2.00/$10.00 MTOKCLAUDE HAIKU 4.5 $1.00/$5.00 MTOKGPT-6 ASTRA $10.00/$50.00 MTOKGPT-5.6 $4.00/$20.00 MTOKGPT-5.6 CYBER $12.50/$75.00 MTOKGPT-5.5 $5.00/$30.00 MTOKGPT-5.4 $2.50/$15.00 MTOKGPT-5.6 TERRA $2.00/$12.00 MTOKGPT-5 MINI $0.25/$2.00 MTOKGPT-5.6 LUNA $0.20/$1.20 MTOKGEMINI 3.1 PRO $2.00/$12.00 MTOKGEMINI 3.6 FLASH $0.75/$3.75 MTOKGEMINI 3.5 FLASH $1.50/$9.00 MTOKGEMINI OMNI FLASH $1.50/$9.00 MTOKGEMINI 3.7 FLASH $0.75/$3.75 MTOKGEMINI 3.8 FLASH $0.75/$3.75 MTOKGEMINI 3 FLASH $0.50/$3.00 MTOKGEMINI 3.5 FLASH LITE $0.30/$2.50 MTOKDEEPSEEK V4 PRO $0.66/$1.98 MTOKDEEPSEEK V4 FLASH $0.22/$0.66 MTOKKIMI K2.6 $0.95/$4.00 MTOK
TOKENBURN_INDEX

THE TICKER · OPENAI · CHECKED DAILY · LAST CHANGE 2026-09-06

GPT-5.6 API PRICING

What OpenAI charges for GPT-5.6, in USD, cross-checked daily against the public pricing dataset. Effective price — if there's a promo running, this is the promo.

INPUT / 1M TOKENS
$4.00
OUTPUT / 1M TOKENS
$20.00
INPUT / 1K TOKENS
$0.0040
OUTPUT MULTIPLE
5.0x
output vs. input price
CACHED INPUT / 1M
$0.400
90% off list input — repeat prefixes only
CACHE WRITE / 1M
$5.00
1.25x list input — the fine print nobody reads
BATCH / 1M (IN / OUT)
$2.00 / $10.00
off-peak / batch, per provider schedule
LONG CONTEXT — ABOVE 272,000 TOKENS
$8.00in (2.0x list)
$30.00out (1.5x list)

OpenAI publishes no separate marginal rate above the line — just this one number, same shape as Google's confirmed whole-request pricing. The scenarios below stay under the line; a large-codebase agent run often doesn't.

NOTE: Priced like gpt-5.6-sol — the mid-high GPT-5.6 tier, one below Cyber.

WHAT THINGS ACTUALLY COST ON GPT-5.6

ScenarioTokens in / out1 run1,000 runs
A typical prompt (300 tokens in)300 / ~900$0.02$19.20
A 20-turn conversation (context re-sent)4,000 / ~4,000$0.10$96.00
Summarizing a 10-page document8,000 / ~3,200$0.10$96.00
One agent run (tools, retries, thinking)2,000 / ~12,000$0.25$248.00
A code review request1,500 / ~3,750$0.08$81.00

Output sizes assume the same editorial ratios as the Burn Calculator (3x for prompts, 6x for agent runs). Estimates, not invoices.

VS. THE REST OF THE BOARD

ModelInput / MTokOutput / MTokSame job costs
GPT-5.6 Cyber OpenAI$12.50$75.003.7x more
Claude Fable 5 Anthropic$10.00$50.002.5x more
Claude Fable 5.1 Anthropic$10.00$50.002.5x more
GPT-6 Astra OpenAI$10.00$50.002.5x more
Claude Opus 5 Anthropic$5.00$25.001.3x more
Claude Opus 4.8 Anthropic$5.00$25.001.3x more
GPT-5.5 OpenAI$5.00$30.001.5x more
GPT-5.4 OpenAI$2.50$15.001.3x less
Claude Sonnet 5 Anthropic$2.00$10.002.0x less
GPT-5.6 Terra OpenAI$2.00$12.001.7x less
Gemini 3.1 Pro Google$2.00$12.001.7x less
Gemini 3.5 Flash Google$1.50$9.002.2x less
Gemini Omni Flash Google$1.50$9.002.2x less
Claude Haiku 4.5 Anthropic$1.00$5.004.0x less
Kimi K2.6 Moonshot AI$0.950$4.004.9x less
Gemini 3.6 Flash Google$0.750$3.755.3x less
Gemini 3.7 Flash Google$0.750$3.755.3x less
Gemini 3.8 Flash Google$0.750$3.755.3x less
DeepSeek V4 Pro DeepSeek$0.660$1.989.7x less
Gemini 3 Flash Google$0.500$3.006.7x less
Gemini 3.5 Flash Lite Google$0.300$2.508.2x less
GPT-5 Mini OpenAI$0.250$2.0010.2x less
DeepSeek V4 Flash DeepSeek$0.220$0.66029.1x less
GPT-5.6 Luna OpenAI$0.200$1.2016.8x less

“Same job” = a 1,000-token prompt with a 3,000-token reply, priced on both models. Cheaper is not always better. It is, however, always cheaper.

PRICE HISTORY

2026-08-23

GPT-5.6 drops to $4 in / $20 out per MTok (-20% in, -33% out).

2026-07-11

GPT-5.6 joins the board at $5.00 in / $30.00 out per MTok.

Source: developers.openai.com/api/docs/pricing · price last confirmed against the daily source on 2026-09-06. The full changelog lives at /pricing/changes.

NOW FIND OUT HOW MUCH OF THAT IS WASTE

GPT-5.6's price is fixed. Your token count isn't. Paste a real prompt into the Burn Calculator and see what it costs on GPT-5.6 — and how much a leaner version would save at 1,000 runs. Or pull the whole board from the ticker.