GPT-5.6 Luna Review (2026): $1/$6 per 1M — Cheapest GPT-5.6 Model
GPT-5.6 Luna: $1/$6 per 1M tokens, fastest GPT-5.6, vision + tools, auto-caching. ChatGPT Free tier. Honest review vs Terra, Sol, GPT-4o, Haiku.
GPT-5.6 Luna is OpenAI's cheapest GPT-5.6 ($1/$6 per 1M) with full vision/tools/code surface. Fastest latency, ChatGPT Free tier. Reasoning caps at high effort. Best for volume inference, chatbots, simple agents. Avoid for complex reasoning, parallel coding, 1M context — use Terra/Sol/GPT-5.5.
GPT-5.6 Luna is OpenAI's efficiency model (July 2026) at $1/$6 per 1M tokens — 20% of Sol cost, 40% of Terra. Fastest in family (~800ms p50, ~90 tok/sec). Full vision + function calling + code execution. Reasoning capped at high effort (no xhigh/max/pro). No ultra multi-agent, no PTC, no explicit caching. ChatGPT Free tier includes rate-limited Luna access. Best for high-volume classification, extraction, chatbots. Avoid for complex reasoning, parallel agentic coding, long-context.
Provider: OpenAI · Family: GPT-5.6
Context window: 200,000 tokens · Max output: 64,000
Input modalities: text, image · Output: text, tool-calls, code
About GPT-5.6 Luna
GPT-5.6 Luna is the efficiency tier of OpenAI's GPT-5.6 family, launched July 9, 2026 for general availability. It delivers the strongest capability-to-cost ratio in the lineup, targeting high-volume inference workloads where per-token cost dominates total economics. Luna sacrifices peak reasoning depth and advanced agentic features for raw speed and affordability. Luna shares the GPT-5.6 architecture family: token efficiency improvements over GPT-5.5, explicit prompt caching with cache breakpoints (30-min TTL), persisted reasoning across turns, and vision + text input with text/tool-calls/code output. It supports reasoning effort levels none, low, medium, and high — xhigh, max, and pro mode are reserved for Terra and Sol. No ultra multi-agent mode, no programmatic tool calling. The model supports 200K context window with 64K max output tokens. Native modalities: text and image input; text, tool-calls, and code output. Vision capabilities present but not benchmarked against dedicated multimodal flagships. No native audio/video I/O. Available on OpenAI API (Responses API, Chat Completions), ChatGPT (all tiers including Free with rate limits), Codex, Microsoft 365 Copilot (explicit selection), and model gateways (OpenRouter, Vercel, Cloudflare, Snowflake, Databricks Mosaic). Knowledge cutoff: June 2026. Pricing: $1.00 per million input tokens, $6.00 per million output tokens. Cached input at $0.125 per million (1.25x base rate, 90% discount). This is 20% of Sol's per-token pricing and 40% of Terra's. No batch discount at launch. Enterprise Zero Data Retention (ZDR) eligible. Data residency: US, EU. SOC2 Type II, ISO 27001, GDPR, HIPAA eligible. Latency: ~800ms p50, ~3000ms p99; ~90 tokens/sec — fastest in GPT-5.6 family.
Pricing
Input $1/M, Output $6/M, Cached input $0.125/M (1.25x base, 90% discount). 20% of Sol price, 40% of Terra price. No batch discount at launch. ChatGPT Free tier includes rate-limited Luna access. Enterprise ZDR eligible. Data residency: US, EU.
Key Features
- Maximum Cost Efficiency: $1/$6 per 1M tokens — 20% of Sol, 40% of Terra. Best token-per-dollar for volume workloads.
- Fastest GPT-5.6 Latency: ~800ms p50, ~90 tokens/sec — 33% faster than Sol, 20% faster than Terra.
- Full Vision + Tool Use: Not stripped down: text/image input, function calling, code execution, structured output all present.
- Auto-Caching + Persisted Reasoning: Automatic prompt caching with 30-min TTL; reasoning.context reuses reasoning across turns.
- ChatGPT Free Tier Access: Only GPT-5.6 model on Free tier (rate-limited) — broadest consumer distribution.
Pros
- Unbeatable cost-efficiency for high-volume inference — $1/$6 with auto-caching makes per-request cost near-zero at scale.
- Fastest latency in family — meets strict SLA requirements for real-time classification, routing, extraction.
- Full capability surface (vision, tools, code) unlike typical 'lite' models that strip features.
- ChatGPT Free tier inclusion = massive distribution and mindshare; upgrade funnel to Plus/Pro/Terra/Sol.
Cons
- Reasoning ceiling at high effort — cannot match Terra/Sol on complex multi-step, novel problems, or deep code analysis.
- No ultra multi-agent, no PTC, no pro mode, no explicit caching — feature subset of Terra with no unique capabilities.
- Distilled model artifacts: overconfidence on edge cases, subtle reasoning gaps vs teacher models.
- Auto-caching only with 1.25x write cost — volatile prefixes burn budget silently.
Benchmarks
- math: 71.5
- mmlu: 88.9
- mmlu pro: 74.8
- aime 2025: 72.1
- arc agi 2: 18.7
- humaneval: 88.4
- live bench: 52.4
- lmarena elo: 1340
- gpqa diamond: 55.3
- lmarena rank: 12
- aider polyglot: 62.1
- swe bench verified: 60.2
- humanitys last exam: 14.2
- artificial analysis intelligence index: 45
- artificial analysis price blended per m: 3.5
- artificial analysis speed tokens per sec: 90
Frequently Asked Questions
What is GPT-5.6 Luna and who built it?
GPT-5.6 Luna is the efficiency tier of OpenAI's GPT-5.6 family, launched July 9, 2026 for general availability. It is a distilled model derived from GPT-5.6 Terra and Sol, optimized for maximum token efficiency and inference speed while retaining the full capability surface: vision + text input, function calling, code execution, structured output. Luna is the fastest and most affordable GPT-5.6 model at $1/$6 per 1M tokens — 20% of Sol's price, 40% of Terra's.
How much does GPT-5.6 Luna cost in 2026?
GPT-5.6 Luna API pricing: $1.00 per 1M input tokens, $6.00 per 1M output tokens. Cached input tokens billed at $0.125 per 1M (1.25x base rate, 90% discount vs full). This is 20% of GPT-5.6 Sol ($5/$30) and 40% of GPT-5.6 Terra ($2.50/$15). For comparison: GPT-4o is $2.50/$10, Claude 3.5 Haiku is $1/$5. ChatGPT Free tier includes rate-limited Luna access. No API free credits; pay-as-you-go from first token. Enterprise ZDR eligible. Data residency: US, EU.
What features does GPT-5.6 Luna have vs Terra and Sol?
Luna has: vision + text input, function calling, code execution, structured output, auto prompt caching (30-min TTL), persisted reasoning (reasoning.context), reasoning effort none/low/medium/high. Luna lacks: xhigh/max/pro reasoning, ultra multi-agent mode, programmatic tool calling (PTC), explicit cache breakpoints. Terra adds: xhigh reasoning, explicit cache breakpoints. Sol adds: ultra multi-agent, PTC, pro mode, max reasoning. All three share 200K context, vision, tools, code execution, auto-caching, persisted reasoning.
Is GPT-5.6 Luna available on ChatGPT Free?
Yes, GPT-5.6 Luna is included in ChatGPT Free tier as of July 9, 2026 — the only GPT-5.6 model available to free users. Access is rate-limited with dynamic limits based on capacity. Free users may also receive GPT-4o or other models depending on load. For guaranteed Luna access with higher limits, ChatGPT Plus ($20/mo) or Pro ($200/mo) required. Plus users get higher Luna limits and Terra access; Pro users get Sol access.
How does GPT-5.6 Luna compare to GPT-4o?
GPT-5.6 Luna vs GPT-4o: Luna is newer architecture (GPT-5.6 family, July 2026 vs GPT-4o May 2024) with better reasoning, token efficiency, and vision quality at similar price ($1/$6 vs $2.50/$10). Luna advantages: 2x token efficiency, stronger benchmarks (SWE-bench 60% vs 45%, GPQA 55% vs 40%), faster latency (800ms vs 1100ms), ChatGPT Free tier. GPT-4o advantages: native audio/video I/O, proven 2-year production track record, wider gateway/integration ecosystem, 128K context (same as Luna). For new projects: Luna. For audio/video multimodal: GPT-4o.
What reasoning modes does GPT-5.6 Luna support?
GPT-5.6 Luna supports four reasoning effort levels: none (fastest, no reasoning tokens), low (light reasoning), medium (balanced default), high (maximum for Luna). It does NOT support xhigh, max, or pro mode — those are exclusive to Terra (xhigh) and Sol (xhigh, max, pro). Setting reasoning.effort=xhigh or max on Luna silently falls back to high. Pro mode errors or falls back. Always set reasoning.effort explicitly to none/low/medium/high.
Does GPT-5.6 Luna have ultra multi-agent mode?
No. Ultra multi-agent mode is exclusive to GPT-5.6 Sol. Luna runs tasks sequentially with standard function calling. It cannot decompose tasks into parallel subagents or synthesize cross-subagent results. For workloads that parallelize (multi-file refactors, parallel test generation, multi-repo analysis), Sol's ultra mode reduces wall-clock time by 40-60%. Luna users must accept sequential execution or route such tasks to Sol.
Who is GPT-5.6 Luna best for and who should avoid it?
Best for: (1) ML Engineers running high-volume inference fleets — $1/$6 with auto-caching makes per-request cost negligible; 90 tok/sec meets SLA. (2) Consumer AI app builders — ChatGPT Free tier Luna access = massive distribution; upgrade funnel to Plus/Pro/Terra/Sol. (3) Cost-optimization leads — price floor that undercuts GPT-4o, Haiku, Flash. (4) Agent fleet operators with bounded-complexity tasks — 5x more agents per dollar vs Terra. Avoid for: (1) Complex multi-step reasoning — reasoning capped at high; xhigh/max/pro unavailable. (2) Frontier agentic coding — no ultra, no PTC; sequential only. (3) 1M+ context document analysis — 200K hard limit; GPT-5.5 has 1M. (4) Audio/video multimodal — vision+text only; GPT-4o, GPT-5.5, Gemini 3.1 have native audio/video.