An Ofox-published guide dated 2026-07-13 lays out the three GPT-5.6 tiers — Sol, Terra, and Luna — and positions Terra as the balanced mid tier at $2.50 input / $15 output per million tokens, halfway between Luna ($1/$6) and Sol ($5/$30). All three tiers share a 1M-token context window and a 128K-token output cap, and The same guide, written by Ofox as the gateway operator exposing Terra via `https://api.ofox.ai/v1`, gives developers a routing pattern: default bulk traffic to Luna or Terra, reserve Sol for the hard tail of requests, and route per-request rather than per-app. It also flags that Terra's Pro/Ultra reasoning modes are o

Ofox's GPT-5.6 Terra catalog listing prices the model at $2.50 per million input tokens and $15 per million output tokens, with $0.25 per million cache reads — exactly half of GPT-5.5 on every line item for a flat 2.0x cost reduction regardless of workload mix. The provider says the model preserves GPT-5.5's 1M-token c The post explicitly warns that OpenAI has not published a Terra-specific coding benchmark, so the half-price value proposition is not paired with a measured parity score. The widely quoted 91.9% Terminal-Bench number is attributed to Sol in Ultra mode, and even flagship Sol is reported to trail Claude Fable 5 on SWE-Be

The CodeRabbit engineering post (July 9, 2026) directly evaluates GPT-5.6 Sol and Terra for coding agents and code review, naming both variants explicitly and positioning them within the GPT-5.6 family alongside a Luna tier. It characterizes Terra as the lower-cost lane suited for scoped implementation and first-pass c The post also contrasts Terra/Sol against competing frontier models it labels "Fable 5" and "Sonnet 5," noting that Sol does not erase those alternatives for architectural judgment or comment quality in some review workflows. The author credits Juan Pablo Flores and Gowtham Kishore Vijay. While the blog is a vendor sou

The llm-stats aggregator page explicitly covers the GPT-5.6 Terra variant and reports concrete benchmark scores tied to this exact model, drawing from OpenAI sources. Notable results include Connectors at 1.00/1 (rank 1), HealthBench Consensus at 0.95/1 (rank 3), Search and Function-Calling at 0.95/1 (rank 1), and GPQA Beyond benchmarks, the page frames GPT-5.6 Terra as a balanced tier in OpenAI's GPT-5.6 family, released in mid-2026, with capability rankings, dataset breakdowns across chat, SVG, games, websites, animations, 3D, audio, and data viz, and a leaderboard standing of 38 of 95 at turn 1. These rankings and category splits

A CodingFleet benchmark and pricing comparison (third-party blog) pits GPT-5.6 Terra against Google's Gemini 3.7 Flash across coding and enterprise-workload evaluations. Terra leads on the pure-coding rows: AA Intelligence Index 57 vs 56, DeepSWE v1.1 69.6% vs 65.3%, Terminal-Bench 2.1 87.4% vs 85.8%, with additional r The same comparison shows Gemini 3.7 Flash winning on enterprise-workflow and document benchmarks (AutomationBench 30.4% vs 23.6%, GDP.pdf 34.0% vs 24.7%, WebDev Arena Elo 1588) while running roughly 3x faster at 340 vs 115 output tokens/sec. Terra's distinctive feature is Programmatic Tool Calling via the Responses AP

AWS published a "What's New" announcement in August 2026 confirming that OpenAI's GPT-5.6 Sol, Terra, and Luna now support 1 million-token context windows on Amazon Bedrock. This makes the same three tiered models available on Bedrock with long context, expanding the install base for Terra beyond the OpenAI API and thi The captured excerpt from the AWS page is dominated by cookie-preference UI boilerplate rather than the announcement body, so the precise date within August 2026 and any region or quota details are not fully visible in the scrape. What is clear is that customers who standardize on Bedrock can now run Terra alongside So