OpenAI

Tier B · Specialist or second-choice models.

GPT-6.1 Terra

A cheap, fast OpenAI option for bounded work, with three working Superbench 2.1 builds out of six: both returning 3D games and one of the four new contracts.

API pricing

USD per million tokens · OpenAI API Standard (≤272K input tokens)
Input
$1
Output
$5
Cached input
$0.05
Context window
1.05M
Pricing details and specifications

Try the active visual tests

Open the generated scene to test it, or compare that prompt with another model.

6 runs

Concourse Divide

Run date: 2026-10-27

Generation provenance: unrecorded

Superbench 2.1 brief · tool-free single-message authoring, no repair turn

Visible agent rerouting, collision avoidance, and recovery from blocked connectivity.

Ember Glider

Run date: 2026-10-28

Generation provenance: 2.1 · repository-run

Superbench 2.1 returning brief · original 1.x run protocol

Sunset gliding journey testing flight energy management, checkpoint flow, and atmospheric scene craft.

Lightkeeper's Atelier

Run date: 2026-10-27

Generation provenance: unrecorded

Superbench 2.1 brief · tool-free single-message authoring, no repair turn

Material differentiation, dynamic light, geometric shadows, and art direction.

Low-Poly Tower Defense

Run date: 2026-10-28

Generation provenance: 2.1 · repository-run

Superbench 2.1 returning brief · original 1.x run protocol

Diorama tower defense testing economy balance, wave design, placement rules, and combat readability.

Wrecklight Salvage

Run date: 2026-10-27

Generation provenance: unrecorded

Superbench 2.1 brief · tool-free single-message authoring, no repair turn

Three-phase game progression with persistent physical cargo and a visible late-stage consequence.

Editorial verdict

Where this model fits

Back to all benchmarks →

GPT-6.1 Terra is OpenAI’s fast, lower-cost GPT-6.1 model. It launched on October 26, 2026 for ChatGPT Work, Codex and the API at $1 input and $5 output per million tokens, half the standard price of GPT-6.1 Sol. Our evidence is one Codex pass on each of the four new Superbench 2.1 contracts, run in parallel the day after launch. Wrecklight Salvage is a working three-phase salvage game, completed by the operator from Phase 1 to an honest dock victory. The other three are failed runs, published as delivered. In Skyhook Cargo the crane’s longest cable cannot reach the landing pad. In Concourse Divide the crowd deadlocks on the first tick. In Lightkeeper’s Atelier the WebGL shader program never links. Under the original repository-run protocol, Terra also built both returning 3D games, and both work: Ember Glider was flown to a 1:51 medal finish against a 3:00 par, and Low-Poly Tower Defense held all ten waves.

Best for

  • High-volume, well-scoped agent and build jobs where cost and speed matter
  • Screen, document and visual-reasoning work
  • Drafting and executor tasks behind a stronger reviewer

Watch out

  • Three of the four new Superbench 2.1 briefs failed: an unreachable pad, a deadlocked crowd and a shader link error
  • Each new brief took about 70 to 95 seconds and 18K to 29K tokens. Terra ships fast but does not check its own logic, and even with tools it stopped before finishing its own validation loop on the 3D games
  • OpenAI itself points complex coding to GPT-6.1 Sol or GPT-6 Astra
  • Provider benchmark numbers are launch-reported and were not reproduced here

Why it is ranked here

  1. OpenAI launched GPT-6.1 Terra on October 26, 2026 at $1 input, $0.05 cached input and $5 output per million tokens, with a 1.05M context window and 128K max output. In Codex the same plan allowance covers far more Terra work than Sol or Astra work.
  2. Four isolated Codex sessions, each holding only one frozen Superbench 2.1 prompt, wrote their four artifacts in 67 to 94 seconds, with no browsing, browser testing or repair turn.
  3. Wrecklight Salvage held up under a full earned run. The vault opened only after two installed cells, the reactor was towed out and through a gate that was really cycling, and victory needed a detached reactor settled in the dock for two seconds.
  4. The other three fail on one logic slip each, not on rendering. Skyhook Cargo caps the cable 0.11 units short of the pad. Concourse Divide puts each team inside the other team’s destination. Lightkeeper’s Atelier declares a uniform with different precision in its two shaders.
  5. With tools and a browser, as the returning-game protocol allows, both 3D builds worked in about two minutes each. The gaps were small: an off-screen ring arrow that only points left and right, and no hover inspection or muzzle flash in the tower game.
  6. B tier is the owner’s editorial placement: use Terra for narrower tasks with a stronger reviewer or a browser check. It is not a score computed from these six runs.

Evidence

2026-11-10

Superbash editorial model ranking

Takeaway: GPT-6.1 Terra is currently placed in Tier B.

The November 2026 editorial roster places GPT-6.1 Terra at rank 16.

Open source →
2026-10-27

GPT-6.1 Terra Superbench 2.1 run record

Takeaway: Six Codex runs, all published as delivered: three working games (both returning 3D games and Wrecklight Salvage) and three failed runs.

The run record lists the protocol, timings, token counts, artifact hashes and per-brief reviewer conclusions.

Open source →

Local benchmark run

All six Superbench 2.1 briefs: four new contracts plus two returning 3D games

Superbash Learn sourceOctober 28, 2026Run record and published artifacts →
Sources, pricing details and methodology

Run identity

Model specifications

Pricing source · checked 2026-10-27 ↗
Status
Available in this Codex session
Codex model ID
gpt-6.1-terra
Run date
October 28, 2026
Access route
OpenAI API, ChatGPT Work and Codex (Plus, Pro, Business, Enterprise and Edu); not in regular ChatGPT chat. The October 27, 2026 Superbench 2.1 runs selected gpt-6.1-terra in Codex CLI 0.163.0; listed API prices are not measured run costs.

Above 272K input tokens the full request is billed at 2x input and cache rates and 1.5x output. Batch and Flex are 50% of Standard; Fast mode is 2x.

Benchmark sources

Skyhook Cargo, Concourse Divide, Lightkeeper’s Atelier and Wrecklight Salvage were each given to a separate GPT-6.1 Terra Codex session, run in parallel. Each session’s directory held only its frozen prompt file, and the session wrote one HTML artifact. No reference implementation, sibling run, other model output or site source was available to any builder. The returning 3D games, Ember Glider and Low-Poly Tower Defense, ran as two parallel repository-run sessions under their original protocol, with tools and a browser allowed. Other models’ runs for those briefs were removed from the checkouts.

GPT-6.1 Terra Superbench 2.1 run record

These are inspectable local artifacts, not provider benchmark scores, passed rubrics or an aggregate rating. Per-artifact reports list which checks passed, which failed and which were not observed. Wrecklight Salvage, Ember Glider and Low-Poly Tower Defense completed earned runs. The other three new briefs are published as failed-run evidence.

Evaluation settings

    Local visual-run methodology and validation records

    Edition coverage

    Recorded means a published artifact, not a behavioral pass. Historical returning results retain their original protocol.

    Superbench 2.1: 6/6 recorded · 0 pending · 0 not run
    • 15 Ember Glider · repository-run · Recorded
    • 14 Low-Poly Tower Defense · repository-run · Recorded
    • 21 Skyhook Cargo · single-prompt · Recorded
    • 22 Concourse Divide · single-prompt · Recorded
    • 23 Lightkeeper's Atelier · single-prompt · Recorded
    • 24 Wrecklight Salvage · single-prompt · Recorded