OpenAIProprietary

GPT-5.6 Terra

Compare this model

A balanced model developed by OpenAI. It achieves performance equivalent to GPT-5.5 at half the cost.

Parameters

Undisclosed

Context Window

License

Proprietary

Release Date

2026-06-26

Japanese Language Capability

High-Quality JP

Multilingual model with strong Japanese language processing capabilities.

API Pricing

Input Price (per 1M tokens)

$2.5

Output Price (per 1M tokens)

$

Billing Mode: standard

Strengths

    Weaknesses

      Use Cases

        Deep Analysis

        AI Coding Agent Index

        77.4

        vs GPT-5.5: 76.4, Claude Fable 5: 77.2

        SWE-Bench Pro

        63.4%

        vs GPT-5.5: 59.4%, Claude Fable 5: 80.3%

        Intelligence Index (v4.1)

        55

        vs GPT-5.5: 54.8, Sol: 58.9, Fable 5: 59.9

        Input Price

        $2.50/1M

        Half of GPT-5.6 Sol ($5/1M)

        Terminal-Bench 2.1

        87.4%

        vs GPT-5.5: 85.6%, Sol: 88.8%

        Agents' Last Exam

        50.4%

        vs GPT-5.5: 46.9%, Sol: 52.7%

        Strengths

        • Offers GPT-5.5-level performance at half the cost for most coding and agentic tasks.
        • Strong reliability (100% success rate in observed benchmarks) and high coding proficiency.
        • Balanced latency (~5.4s) and cost, making it ideal for high-volume production workloads.

        Weaknesses

        • Trails on the hardest math benchmarks (FrontierMath Tier 4: 68.3% vs GPT-5.5: 72.5%).
        • Perceived by some as a potential distillation from Sol rather than a pure architectural advancement.
        • Weak long-context recall compared to higher tiers (MRCR: 89.6% vs Sol's 91.5% and Luna's 41.3%).

        Competitor Comparison

        ModelArenaSWEGPQAPrice
        GPT-5.5N/A59.4%93.6%$5/$30
        Claude Fable 5N/A80.3%92.6%$10/$50
        GPT-5.6 SolN/A64.6%94.6%$5/$30

        GPT-5.6 Terra is the balanced, mid-tier model in OpenAI's GPT-5.6 family, launched on July 9, 2026. Positioned as a "sensible default," it delivers performance competitive with the previous generation's flagship, GPT-5.5, at exactly half the token cost ($2.50/$15 per million input/output tokens). Its primary value proposition is cost-effective intelligence for everyday professional, coding, and agentic workflows, making it a compelling upgrade path for teams currently using GPT-5.5 or other mid-tier models. OpenAI positions Terra for "everyday work," where Sol is overkill and Luna is underpowered, creating a clear three-tier product strategy for different workload intensities.

        Terra's benchmark performance demonstrates its balanced nature. It nearly matches GPT-5.5 on key agentic and coding benchmarks (e.g., Agents' Last Exam, Terminal-Bench) while significantly outperforming it on cost-efficiency. Independent analysis from Artificial Analysis confirms it defines a new Pareto frontier of intelligence versus cost for the mid-tier. However, it cedes the absolute ceiling to the flagship Sol on the most complex tasks and trails on certain pure-math evaluations. The model also introduces OpenAI's first cache-write pricing, aligning with Anthropic's approach for more predictable costs in applications with repeated prefixes.

        The launch sparked discussion about whether Terra represents a genuine capability step or is primarily a distilled, cost-optimized version of Sol. While its benchmark gains over GPT-5.5 are real on most agentic and coding tasks, skepticism remains in the developer community. Regardless, Terra is strategically significant as it allows OpenAI to compete aggressively on cost-performance across the entire workload spectrum, from high-volume Luna pipelines to frontier Sol tasks, solidifying its position in the market against rivals like Anthropic and Google.

        Analysis generated: 2026-07-17