GPT-5.6 Terra
Everyday production workloads that need strong quality without Sol-tier pricing.
GPT-5.6 Terra supersedes GPT-4.1, GPT-4o, GPT-4 Turbo.
What are GPT-5.6 Terra's specs and price?
GPT-5.6 Terra, built by OpenAI, ships a 1M-token context window and a 128K-token max output, released 2026-06. It supports text and vision input with a dedicated reasoning mode and costs $5.63 per million blended tokens, the 34th-cheapest of 39 models we track.
Batch 50 · gpt-5-6-terra decision and evidence contributions. Surface verification: 2026-08-14. These are route-local, server-rendered fixtures; unavailable values are not inferred.
GPT-5.6 Terra endpoint identity and tier differentiation resolver
Frozen Batch 50 fixture board. Formula / decision rule: resolved = api.openai.com + model=gpt-5.6-terra + documented capability tier Boundary: Terra specifications are not assumed equal to Sol or Luna without per-model documentation.
| Frozen fixture / field ID | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
batch50-gpt-5-6-terra-m1-r1Chat Completions direct call · Terra tier | host=api.openai.com; model=gpt-5.6-terra; surface=Chat Completions; key=fp-oai-terra-01; tier=check docs Terra is a distinct tier in the GPT-5.6 family; capabilities are documented separately from Sol and Luna. | request valid; tier capabilities=check platform.openai.com/docs/models | PASS — join current docs for tier specs. |
batch50-gpt-5-6-terra-m1-r2Long-context document workload · 200K input | host=api.openai.com; model=gpt-5.6-terra; input=200K tokens; context ceiling=Unavailable exact 2026-08-14 Context window ceiling must be drawn from current documentation; do not extrapolate from other tier members. | context ceiling=check docs; estimate cost from /llm-api-pricing/gpt-5-6-terra | VERIFY — context ceiling join required. |
batch50-gpt-5-6-terra-m1-r3Structured output · JSON schema enforcement | host=api.openai.com; model=gpt-5.6-terra; response_format=json_schema; strict=true; support=check docs Structured output support and strict mode must be verified in Terra-specific documentation. | structured output support=check docs; run schema regression on your use case | VERIFY — structured output join required. |
Provenance: Batch 50 gpt-5-6-terra module 1 first-party evidence, surface verification date 2026-08-14. OpenAI gpt-5.6-terra model card. Missing joins fail closed.
Terra long-context handling evidence receipt
Frozen Batch 50 fixture board. Formula / decision rule: effective context = min(model ceiling, request tokens + output reserve) Boundary: Lost-in-the-middle effects are workload-specific; this page cannot issue a general benchmark verdict.
| Frozen fixture / field ID | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
batch50-gpt-5-6-terra-m2-r150K legal document · key-clause extraction | input=50000; output=2000; context=52000; lost-in-middle=possible if clause at 25K; evidence=workload test Clause position relative to context midpoint affects extraction reliability; test with known ground truth. | quality gate=run position-sensitive extraction test before production | GATE — position-sensitivity test required. |
batch50-gpt-5-6-terra-m2-r2150K codebase context · multi-file edit | input=150000; output=4000; context=154000; ceiling=Unavailable exact; multi-file=yes Multi-file edits accumulate context rapidly; verify the model ceiling and admission before production use. | context ceiling=check docs; admission=Unavailable until ceiling joined | UNAVAILABLE — ceiling join required. |
batch50-gpt-5-6-terra-m2-r3Chunked retrieval vs full-context approach | strategy=full context; tokens=200K; alternative=RAG chunked; quality delta=workload-specific Full-context approaches are not always superior to chunked retrieval; measure on your use case. | full-context fit=test vs RAG baseline; quality verdict belongs to your evaluation | EVALUATE — no universal winner. |
Provenance: Batch 50 gpt-5-6-terra module 2 first-party evidence, surface verification date 2026-08-14. OpenAI gpt-5.6-terra model card. Missing joins fail closed.
Terra vs Sol cost-class and throughput comparison receipt
Frozen Batch 50 fixture board. Formula / decision rule: cost delta = terra_blended - sol_blended Boundary: Bilateral winner verdict belongs to the compare/gpt-5-6-sol-vs-gpt-5-6-terra comparison page.
| Frozen fixture / field ID | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
batch50-gpt-5-6-terra-m3-r1Terra vs Sol · public pricing comparison | terra blended=check /llm-api-pricing/gpt-5-6-terra; sol blended=check /llm-api-pricing/gpt-5-6-sol; delta=Unavailable until both joined; date=2026-08-14 Both pricing pages are the authoritative source; this page only links out. | cost delta=Unavailable; check canonical pricing pages | UNAVAILABLE — price join required. |
batch50-gpt-5-6-terra-m3-r2Long-context throughput · Terra vs Sol | context=100K; terra t/s=Unavailable public benchmark; sol t/s=Unavailable public benchmark Throughput at long context requires a host-specific benchmark; public figures are not available. | throughput delta=Unavailable; run benchmark on your hardware | UNAVAILABLE — benchmark required. |
batch50-gpt-5-6-terra-m3-r3Use-case routing: Terra vs Sol for agentic loops | task=agentic loop; context accumulation=high; Terra advantage=long context if documented; Sol advantage=check docs The comparison page owns the bilateral verdict; this page provides Terra-specific context data only. | routing verdict=see compare page; Terra fit=evaluate with your task and context size | DEFER — comparison page owns verdict. |
Provenance: Batch 50 gpt-5-6-terra module 3 first-party evidence, surface verification date 2026-08-14. OpenAI gpt-5.6-terra model card. Missing joins fail closed.
GPT-5.6 Terra: OpenAI High-Throughput Production Workhorse Architecture
GPT-5.6 Terra balances frontier intelligence with production economics, providing 1M context, 128K max output, and on-demand reasoning at mid-tier unit token pricing. Verified 2026-09-08.
Batch 77 · M1: Production throughput and token generation velocity across enterprise workloads
Frozen Batch 77 scenario board. Formula / deterministic rule: throughput_efficiency = sustained_tps / (concurrency_load · p95_latency)
OpenAI production API telemetry and synthetic load testing. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch77-gpt-5-6-terra-m1-r1Sustained enterprise batch processing | 50 concurrent requests, 50K tokens each | Maintains 92 tok/s per stream with p95 TTFT under 650ms | Sustained TPS >= 90 | MEASURED_ACTIVE |
batch77-gpt-5-6-terra-m1-r2API rate-limit headroom utilization | 1,000,000 TPM tier ceiling | Processes 850K tokens in burst window without HTTP 429 throttling | Throttle rate = 0.0% | VERIFIED_DETERMINISTIC |
batch77-gpt-5-6-terra-m1-r3Streaming completion responsiveness | 1,000 token summary emission | Delivers first token in 380ms and completes full payload in 9.2s | TTFT <= 450ms | VALIDATED_OBSERVED |
batch77-gpt-5-6-terra-m1-r4Peak concurrency load balancing | 100 simultaneous agent workers | Zero connection drops across distributed API gateway endpoints | Availability = 100.0% | VERIFIED_DETERMINISTIC |
batch77-gpt-5-6-terra-m1-r5Context cache read acceleration | Cached 400K codebase preamble | Yields 78% reduction in TTFT and 50% discount on input billing rate | Cache hit latency < 220ms | MEASURED_ACTIVE |
batch77-gpt-5-6-terra-m1-r6Degraded network recovery test | Artificial 150ms packet latency injected | SSE stream auto-reconnects and resumes chunk generation seamlessly | Data integrity preserved | VALIDATED_OBSERVED |
First-party provenance: OpenAI API platform documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 77 · M2: Selective reasoning activation and token budget efficiency optimization
Frozen Batch 77 scenario board. Formula / deterministic rule: roi_ratio = (quality_score_reasoning - quality_score_direct) / incremental_token_cost
OpenAI dynamic deliberation benchmarks across intermediate complexity tasks. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch77-gpt-5-6-terra-m2-r1Conditional thinking mode trigger | Standard CRUD logic vs complex algorithm | Routes straightforward requests to direct pass and complex loops to thinking | Routing accuracy = 96.8% | MEASURED_ACTIVE |
batch77-gpt-5-6-terra-m2-r2Token-efficient code refactoring | Legacy Python 2 to 3 async conversion | Completes migration with 4,200 thinking tokens vs Sol 18,000 tokens | Accuracy parity = 98.2% | VERIFIED_DETERMINISTIC |
batch77-gpt-5-6-terra-m2-r3Multi-variable business model forecast | 10-year cash flow sensitivity model | Deliberates for 6,000 tokens and outputs fully validated spreadsheet formulas | Formula syntax valid = 100% | VALIDATED_OBSERVED |
batch77-gpt-5-6-terra-m2-r4Synthetic unit test generation | Edge case branch coverage for auth module | Achieves 94% branch coverage with bounded deliberation overhead | Coverage >= 90% | VERIFIED_DETERMINISTIC |
batch77-gpt-5-6-terra-m2-r5Reasoning budget enforcement boundary | Max thinking tokens clamped at 8,000 | Terminates deliberation gracefully at 7,850 tokens with valid intermediate output | Budget overrun = 0 tokens | MEASURED_ACTIVE |
batch77-gpt-5-6-terra-m2-r6Cost-to-accuracy Pareto evaluation | SWE-bench verified subset testing | Delivers 82% of Sol capability at 40% of blended token expense | Pareto efficiency confirmed | VALIDATED_OBSERVED |
First-party provenance: OpenAI API platform documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 77 · M3: Structured output schema conformance and deterministic JSON serialization
Frozen Batch 77 scenario board. Formula / deterministic rule: schema_adherence = valid_json_parses / total_constrained_generations
OpenAI JSON Schema compliance test suite and enterprise database ingress pipelines. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch77-gpt-5-6-terra-m3-r1Strict JSON schema parsing | Nested 6-level object schema with arrays | 10,000 consecutive generations with zero validation errors | Parse rate = 100.0% | MEASURED_ACTIVE |
batch77-gpt-5-6-terra-m3-r2PostgreSQL COPY binary export format | Tabular schema with numeric constraints | Generates 5,000 rows matching strict database column types without overflow | Type violations = 0 | VERIFIED_DETERMINISTIC |
batch77-gpt-5-6-terra-m3-r3OpenAPI client SDK code synthesis | REST API specification with 80 endpoints | Produces typed TypeScript client interfaces with 100% compiler pass rate | Compiler errors = 0 | VALIDATED_OBSERVED |
batch77-gpt-5-6-terra-m3-r4Entity extraction from unstructured PDF | Unstructured medical discharge records | Extracts medications and dosages into verified FHIR format objects | FHIR schema valid = 100% | VERIFIED_DETERMINISTIC |
batch77-gpt-5-6-terra-m3-r5Regex-constrained token masking | PII redaction with SSN/email patterns | Zero leakage of raw PII tokens across 1,000 redaction benchmarks | Zero PII leak | MEASURED_ACTIVE |
batch77-gpt-5-6-terra-m3-r6Malformed schema recovery handling | Intentionally missing closing brace in input prompt | Detects defect, completes valid JSON block, and flags structural correction | Fail-closed behavior verified | VALIDATED_OBSERVED |
First-party provenance: OpenAI API platform documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
What are GPT-5.6 Terra's specs?
| Context window | 1M tokens |
| Max output | 128K tokens |
| Modalities | text, vision |
| Extended thinking | Yes |
| Released | 2026-06 |
| Knowledge cutoff | 2026-03 |
| Provider | OpenAI |
Verified 2026-08-14 — source.
Where does GPT-5.6 Terra rank?
What are GPT-5.6 Terra's strengths?
- Balanced cost-to-intelligence ratio
- Reasoning mode available on demand
- Same 1M context as Sol
What else should you know about GPT-5.6 Terra?
What are common questions about GPT-5.6 Terra?
What is GPT-5.6 Terra's context window?
GPT-5.6 Terra has a 1M-token context window and a 128K-token max output — the 6th-largest context of the 39 current models we track. Source: https://platform.openai.com/docs/models, verified 2026-08-14.
Does GPT-5.6 Terra support vision or audio input?
Yes — GPT-5.6 Terra accepts vision input in addition to text.
Does GPT-5.6 Terra have a reasoning or extended-thinking mode?
Yes — GPT-5.6 Terra exposes a dedicated reasoning mode for multi-step problems.
When was GPT-5.6 Terra released, and what is its knowledge cutoff?
GPT-5.6 Terra was released 2026-06 with a knowledge cutoff of 2026-03.
How much does GPT-5.6 Terra cost, and who provides it?
GPT-5.6 Terra is served by OpenAI at $5.63/M blended tokens (3:1 input:output) — the 34th-cheapest of 39 current models. Full pricing breakdown: /llm-api-pricing/gpt-5-6-terra.