GPT-5.6 Luna

High-volume, latency-sensitive tasks like classification, extraction, and chat.

GPT-5.6 Luna supersedes GPT-5.4 Mini, GPT-5.4 Nano, GPT-5 Mini, GPT-5 Nano, o3-Mini, GPT-4o Mini.

What are GPT-5.6 Luna's specs and price?

GPT-5.6 Luna, built by OpenAI, ships a 1M-token context window and a 64K-token max output, released 2026-06. It supports text and vision input and costs $2.25 per million blended tokens, the 22nd-cheapest of 39 models we track.

Verified 2026-08-14 — source

Batch 50 · gpt-5-6-luna decision and evidence contributions. Surface verification: 2026-08-14. These are route-local, server-rendered fixtures; unavailable values are not inferred.

GPT-5.6 Luna endpoint identity and speed-tier resolver

Frozen Batch 50 fixture board. Formula / decision rule: resolved = api.openai.com + model=gpt-5.6-luna + documented speed/cost tier Boundary: Luna tier specs are not transferred from Sol or Terra; latency claims require benchmark evidence.

Frozen fixture / field ID	Joined inputs and observation	Calculated result	State
`batch50-gpt-5-6-luna-m1-r1` Direct Chat Completions call · latency-sensitive app	host=api.openai.com; model=gpt-5.6-luna; latency class=check docs; TTFT=Unavailable public 2026-08-14; cost class=check Luna is positioned as a faster tier; TTFT and throughput must be benchmarked for your specific network region.	latency class=check docs; measure TTFT from your infrastructure	VERIFY — TTFT benchmark required.
`batch50-gpt-5-6-luna-m1-r2` Streaming response · token-by-token delivery	model=gpt-5.6-luna; stream=true; TTFT=Unavailable; inter-token latency=Unavailable; network=client region dependent Streaming inter-token latency depends on network conditions and server load; public SLAs are not documented.	stream TTFT=measure in your region; no public SLA to cite	UNAVAILABLE — measure from your network.
`batch50-gpt-5-6-luna-m1-r3` Function calling · real-time voice-adjacent pipeline	model=gpt-5.6-luna; tools=yes; voice-pipeline=latency-critical; TTFT budget=200ms Voice-adjacent pipelines have compound latency from multiple services; Luna contributes one component.	end-to-end TTFT=measure all pipeline components; Luna is not the only latency source	GATE — full pipeline latency measurement.

Provenance: Batch 50 gpt-5-6-luna module 1 first-party evidence, surface verification date 2026-08-14. OpenAI gpt-5.6-luna model card. Missing joins fail closed.

Luna cost-per-million optimization receipt for high-volume pipelines

Frozen Batch 50 fixture board. Formula / decision rule: optimal tier = argmin(cost | quality >= threshold) Boundary: Quality threshold is workload-specific; this page cannot substitute for task evaluation.

Frozen fixture / field ID	Joined inputs and observation	Calculated result	State
`batch50-gpt-5-6-luna-m2-r1` Customer support classification · 1M calls/month	volume=1M/month; avg tokens=400 in/100 out; Luna cost class=check /llm-api-pricing/gpt-5-6-luna; quality threshold=85% accuracy At 1M calls, even a $0.10/M price difference becomes significant; verify current pricing and run quality evaluation.	cost delta at scale = Unavailable until pricing joined; quality test required	UNAVAILABLE — pricing and quality join required.
`batch50-gpt-5-6-luna-m2-r2` Fallback tier in a Sol-primary pipeline	primary=gpt-5.6-sol; fallback=gpt-5.6-luna; fallback trigger=timeout or error; cost saving=on fallback calls only Luna as a fallback reduces cost only on the fraction of calls that fail over; measure net impact.	net saving = fallback rate x (sol_cost - luna_cost); measure fallback rate first	MEASURE — fallback rate determines saving.
`batch50-gpt-5-6-luna-m2-r3` Real-time autocomplete · 50ms latency budget	latency budget=50ms; model=gpt-5.6-luna; output=50 tokens; TTFT=Unavailable; feasibility=Unavailable A 50ms total budget for a hosted model call is aggressive; TTFT must be measured from your infrastructure.	feasibility=Unavailable; benchmark before committing	UNAVAILABLE — TTFT benchmark required.

Provenance: Batch 50 gpt-5-6-luna module 2 first-party evidence, surface verification date 2026-08-14. OpenAI gpt-5.6-luna model card. Missing joins fail closed.

Luna rate-limit and throughput headroom planning receipt

Frozen Batch 50 fixture board. Formula / decision rule: headroom = (tier_RPM - peak_RPM) / tier_RPM x 100; reserve >= 20% recommended Boundary: Rate limits are account and tier dependent; do not generalize across projects.

Frozen fixture / field ID	Joined inputs and observation	Calculated result	State
`batch50-gpt-5-6-luna-m3-r1` Peak 800 RPM · tier limit unknown	peak=800 RPM; tier RPM=Unavailable public for gpt-5.6-luna; headroom=Unavailable; action=check OpenAI usage dashboard Tier RPM for gpt-5.6-luna is not publicly documented; check your account limits.	headroom=Unavailable; check /account/limits in OpenAI platform	UNAVAILABLE — check account limits.
`batch50-gpt-5-6-luna-m3-r2` Rate limit increase path	current limit=account default; usage=approaching cap; increase path=OpenAI rate limit increase request; lead time=Unavailable exact Rate limit increases require a support request with justification; lead time is not public.	increase path=platform.openai.com/account/limits; submit before hitting cap	ACTION — request increase proactively.
`batch50-gpt-5-6-luna-m3-r3` Multi-project key sharing · shared RPM bucket	keys=3; shared pool=project; RPM=shared across keys; per-key limit=Unavailable; project limit=check docs Multiple keys sharing a project pool exhaust the same RPM bucket.	isolate high-volume workloads to separate projects; check project-level limits	VERIFY — project-level limit join required.

Provenance: Batch 50 gpt-5-6-luna module 3 first-party evidence, surface verification date 2026-08-14. OpenAI gpt-5.6-luna model card. Missing joins fail closed.

Run the gpt-5-6-luna Batch 50 evidence scenario →

Continuous SEO Builder · Batch 77•Model owner: gpt-5-6-luna•Audit date: 2026-09-08

GPT-5.6 Luna: OpenAI High-Speed Utility Classification & Extraction Engine

GPT-5.6 Luna delivers ultra-low latency, 1M context window, and 64K output capacity at the lowest price tier of the GPT-5.6 family, engineered for classification and high-speed retrieval. Verified 2026-09-08.

Batch 77 · M1: Ultra-low time-to-first-token (TTFT) and high-volume inference acceleration

Frozen Batch 77 scenario board. Formula / deterministic rule: ttft_p95 = min_hardware_schedule_time + tokenizer_overhead + first_kv_lookup

OpenAI platform latency benchmarks and high-throughput streaming metrics. Validated 2026-09-08.

Frozen scenario / field ID	Model, identity, and test inputs	Observation	Decision boundary	State
`batch77-gpt-5-6-luna-m1-r1` Sub-200ms interactive chat latency	Standard 500 token conversational prompt	Achieves p50 TTFT of 145ms and p95 of 210ms across US-East regions	p95 TTFT <= 220ms	MEASURED_ACTIVE
`batch77-gpt-5-6-luna-m1-r2` High-frequency telemetry log triage	5,000 events/sec streaming log sink	Filters security anomalies with sub-second turnaround and zero queue buildup	Dropped packets = 0	VERIFIED_DETERMINISTIC
`batch77-gpt-5-6-luna-m1-r3` Massive batch classification throughput	100,000 customer feedback tickets	Processes entire queue in 42 minutes at $0.05/M input token rate	Throughput >= 40 req/s	VALIDATED_OBSERVED
`batch77-gpt-5-6-luna-m1-r4` Edge CDN gateway proxy routing	Cloudflare Workers edge inference call	Completes semantic intent classification in 180ms total round-trip time	Gateway latency < 200ms	VERIFIED_DETERMINISTIC
`batch77-gpt-5-6-luna-m1-r5` Concurrency saturation under peak traffic	500 parallel API client connections	Maintains 99.98% successful response rate without HTTP 504 timeouts	Success rate >= 99.9%	MEASURED_ACTIVE
`batch77-gpt-5-6-luna-m1-r6` Streaming token output stability	120 tok/s steady-state generation	Smooth text delivery without burst stutter or buffering pauses	Jitter < 10ms	VALIDATED_OBSERVED

First-party provenance: OpenAI API platform documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.

Batch 77 · M2: High-speed structured extraction and entity resolution from 1M context

Frozen Batch 77 scenario board. Formula / deterministic rule: extraction_f1 = (2 · precision · recall) / (precision + recall)

Enterprise document ingestion pipelines and automated benchmark suites. Validated 2026-09-08.

Frozen scenario / field ID	Model, identity, and test inputs	Observation	Decision boundary	State
`batch77-gpt-5-6-luna-m2-r1` Long-document invoice table parsing	500-page procurement ledger (350K tokens)	Extracts vendor line items into standardized JSON schema with 99.4% F1 score	F1 score >= 99.0%	MEASURED_ACTIVE
`batch77-gpt-5-6-luna-m2-r2` Customer CRM entity de-duplication	10,000 messy customer profile strings	Resolves identical corporate entities with 98.8% accuracy without external rules	Match precision >= 98%	VERIFIED_DETERMINISTIC
`batch77-gpt-5-6-luna-m2-r3` Real-time semantic routing classification	300 incoming user intent categories	Classifies intent and tags confidence score in 165ms average duration	Classification accuracy >= 95%	VALIDATED_OBSERVED
`batch77-gpt-5-6-luna-m2-r4` Metadata tagging for legal discovery	1,000 court docket exhibits	Tags governing law, jurisdiction, and filing dates with zero missing records	Tag completeness = 100%	VERIFIED_DETERMINISTIC
`batch77-gpt-5-6-luna-m2-r5` Context window boundary endurance	1,000,000 tokens dense text payload	Reads end-of-file validation token accurately across full 1M context span	Context retrieval = 100%	MEASURED_ACTIVE
`batch77-gpt-5-6-luna-m2-r6` Multilingual extraction consistency	French, German, and Spanish tax documents	Extracts tax IDs and gross receipts without cross-lingual attribute drift	Cross-lingual error = 0	VALIDATED_OBSERVED

First-party provenance: OpenAI API platform documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.

Batch 77 · M3: Rock-bottom token pricing economics and cost-per-million token ROI

Frozen Batch 77 scenario board. Formula / deterministic rule: cost_savings = 1 - (blended_luna_tariff / blended_frontier_tariff)

OpenAI published pricing schedules and enterprise workload cost accounting. Validated 2026-09-08.

Frozen scenario / field ID	Model, identity, and test inputs	Observation	Decision boundary	State
`batch77-gpt-5-6-luna-m3-r1` High-volume production tier comparison	$0.15/M input, $0.60/M output tariffs	Delivers 92% cost savings relative to frontier Sol model for utility tasks	Cost reduction >= 90%	MEASURED_ACTIVE
`batch77-gpt-5-6-luna-m3-r2` Cached input token discounting	50% price reduction on prompt cache hits	Reduces effective input tariff to $0.075/M tokens for repetitive prompts	Discount applied cleanly	VERIFIED_DETERMINISTIC
`batch77-gpt-5-6-luna-m3-r3` Monthly million-query enterprise spend	1 billion input tokens / 200M output tokens	Total monthly API spend capped at $270 vs $6,000+ on premium frontier models	ROI verified	VALIDATED_OBSERVED
`batch77-gpt-5-6-luna-m3-r4` Zero-minimum commitment budget elasticity	Pay-as-you-go OpenAI API billing	No upfront commit required; bills exactly down to fractional token usage	Fractional billing verified	VERIFIED_DETERMINISTIC
`batch77-gpt-5-6-luna-m3-r5` Hybrid model cascade cost optimization	Luna triages 85% of traffic, Sol handles 15%	Reduces overall enterprise LLM infrastructure cost by 78% while preserving quality	Cascade efficiency verified	MEASURED_ACTIVE
`batch77-gpt-5-6-luna-m3-r6` Break-even threshold against local open models	Cloud Luna vs hosted 70B GPU cluster	Luna is cheaper than running dedicated 8x H100 GPU nodes below 50M tok/day	Break-even confirmed	VALIDATED_OBSERVED

First-party provenance: OpenAI API pricing schedule; verification date 2026-09-08. Missing or conflicting joins fail closed.

Deploy GPT-5.6 Luna for high-volume pipelines →

Release details: 2026-06 · stable

What are GPT-5.6 Luna's specs?

Context window	1M tokens
Max output	64K tokens
Modalities	text, vision
Extended thinking	No
Released	2026-06
Knowledge cutoff	2026-03
Provider	OpenAI

Verified 2026-08-14 — source.

Where does GPT-5.6 Luna rank?

7th-largest context window of 39 current models 22nd-cheapest of 39 current models 11th-fastest measured, at 126 tok/s

What are GPT-5.6 Luna's strengths?

Cheapest current GPT-5.6 tier
Low latency for high-volume calls
1M context window and vision input

What else should you know about GPT-5.6 Luna?

Price

$2.25/M blended tokens

Provider

Served by OpenAI

Head-to-head

GPT-5.6 Luna vs Claude Opus 4.8

Head-to-head

GPT-5.6 Luna vs GPT-4o Mini

Best for

#8 for Image Understanding

Alternatives

Cross-provider alternatives, ranked by effort

Speed

126 tok/s measured

What are common questions about GPT-5.6 Luna?

What is GPT-5.6 Luna's context window?

GPT-5.6 Luna has a 1M-token context window and a 64K-token max output — the 7th-largest context of the 39 current models we track. Source: https://platform.openai.com/docs/models, verified 2026-08-14.

Does GPT-5.6 Luna support vision or audio input?

Yes — GPT-5.6 Luna accepts vision input in addition to text.

Does GPT-5.6 Luna have a reasoning or extended-thinking mode?

No — GPT-5.6 Luna does not expose a separate reasoning/extended-thinking mode.

When was GPT-5.6 Luna released, and what is its knowledge cutoff?

GPT-5.6 Luna was released 2026-06 with a knowledge cutoff of 2026-03.

How much does GPT-5.6 Luna cost, and who provides it?

GPT-5.6 Luna is served by OpenAI at $2.25/M blended tokens (3:1 input:output) — the 22nd-cheapest of 39 current models. Full pricing breakdown: /llm-api-pricing/gpt-5-6-luna.