Home › AI Benchmark › GPT-5.6 Terra
OpenAI released on Jul 9, 2026 API access

GPT-5.6 Terra

GPT-5.6 Terra is the middle tier of the GPT-5.6 family, released by OpenAI on Jul 9, 2026 between Sol and Luna; OpenAI compares it to the old mini tier. It costs US$ 2 per million input tokens and US$ 12 per million output, 20% less than at launch. The GPT-6 family has no Terra, and GPT-6.1 Sol now costs the same for input and less for output and cache. No retirement date has been published.

#12 of 267 in the ECI ranking
159.43 ECI, the Epoch AI capability index
1,515 Arena ELO, human preference · #35
77.9 LiveBench, overall average from 0 to 100 · #18
No votes yet. Rate it.
Spec sheet

GPT-5.6 Terra in official numbers

Data published by the maker in its documentation and announcement. The source of each item is at the end of the page.

Maker
OpenAI
API identifier
gpt-5.6-terra
Release
Jul 9, 2026, with GPT-5.6 Sol and GPT-5.6 Luna
Input
US$ 2 per million tokens (was US$ 2.50 until Jul 30, 2026)
Output
US$ 12 per million tokens (was US$ 15)
Cache read
US$ 0.20 per million (10% of input)
Cache write
US$ 2.50 per million (1.25x input)
Above 272 thousand input tokens
US$ 4 input, US$ 0.40 cache and US$ 18 output, for the whole request
Batch and Flex
US$ 1 input and US$ 6 output
Fast mode
US$ 4 and US$ 24
Context window
1,050,000 tokens (up to 922 thousand input)
Max output
128 thousand tokens
Knowledge cutoff
Feb 16, 2026
Reasoning
effort none to max, default medium
Equivalent to
the mini tier of earlier GPT-5 families, according to OpenAI
Where it runs
OpenAI API, ChatGPT Work, Codex and Amazon Bedrock (openai.gpt-5.6-terra)
Risk
High in cybersecurity and biology on the Preparedness scale, like Sol
Retirement
no date published as of Oct 1, 2026
Analysis

Where GPT-5.6 Terra does well, where it falls behind and who it is for

Where it does well

  • Close to Sol: 50.4% on Agents' Last Exam and 69.6% on DeepSWE, against 52.7% and 72.7% for GPT-5.6 Sol, at half the input price.
  • Above GPT-5.5: 77.4 points on the Artificial Analysis Coding Agent Index, against 76.4 for GPT-5.5 and 77.2 for Claude Fable 5, in OpenAI's table.
  • Long context: 89.6% on the MRCR v2 test from 256 thousand to 512 thousand tokens, close to Sol's 91.5% and more than double GPT-5.6 Luna's 41.3%.
  • Fewer harmful agent actions: 13.5% misaligned outcomes in simulated work environments, without a confirmation policy, against 18.8% for Sol and 19.7% for Luna, in the GPT-6 Astra system card.
  • Prompt injection: 0.946 on the search and function-calling evaluation in the GPT-5.6 system card, the family's highest score (Sol 0.910).

Where it falls behind

  • Pricier than the next generation's Sol: GPT-6.1 Sol costs the same for input, US$ 10 against US$ 12 for output and half for cache.
  • Computer use: 50.2% on OSWorld 2.0, against 62.6% for Sol and 54.8% for Claude Opus 4.8.
  • Hard math: 68.3% on FrontierMath Tier 4, below even GPT-5.5 (72.5%).
  • Hard software engineering: 63.4% on SWE-Bench Pro, against 80% for Claude Fable 5 in the same table.
  • No successor: there is no GPT-6 Terra, and OpenAI's model selection guide no longer mentions Terra.

Who it makes sense for

  • Integrations already running on gpt-5.6-terra and validated: there is no shutdown date, and OpenAI promises at least 6 months' notice.
  • Teams leaving gpt-5-mini, o4-mini or gpt-3.5-turbo: it is the target the official replacement table names, with shutdowns on Oct 23, 2026 and Dec 11, 2026.
  • Documents too long for GPT-5.6 Luna: on the needle test from 256 thousand to 512 thousand tokens, Terra scores more than double.
  • Codex on ChatGPT's Free and Go plans, which have access to Terra, according to OpenAI's help center.

When to pick another model

  • New projects: GPT-6.1 Sol costs the same or less on every line of the price table and is the middle tier OpenAI recommends today.
  • High-volume classification, extraction and summaries: GPT-6 Luna costs US$ 0.10 and US$ 0.50, one twentieth of Terra's input price.
  • Interface automation and computer use: on OSWorld 2.0, Sol scored 62.6% against Terra's 50.2%.
  • Tasks where GPT-5.6 Luna ties: on Agents' Last Exam, Luna scores 50.3% against Terra's 50.4%, at a tenth of the price.

What OpenAI published at launch

Terra came out on Jul 9, 2026 as the balanced tier of the GPT-5.6 family. OpenAI described it as a lower-cost model with performance competitive with GPT-5.5, and the documentation says it roughly corresponds to the old mini tier.

The launch price was US$ 2.50 input and US$ 15 output. On Jul 30 it fell 20%, to US$ 2 and US$ 12, on the same day Luna became 80% cheaper.

In the announcement table, Terra does not lead any of the eleven tests below, but beats GPT-5.5 on nine. According to OpenAI, it lands just above Claude Fable 5 on the Artificial Analysis coding agent index, at about a quarter of the estimated cost.

GPT-5.6 Terra in the figures OpenAI published
GPT-5.6 Terra
OpenAI		GPT-5.6 Sol
OpenAI		GPT-5.5
OpenAI		Claude Fable 5
Anthropic
Agents' Last Exam
professional agent work		 50.4%52.7%46.9%40.5%
GDPval-AA v2
office work, Elo		 1,5931,747.81,493.71,759.6
AA Coding Agent Index v1.1
coding agent		 77.48076.477.2
SWE-Bench Pro
software engineering		 63.4%64.6%59.4%80%
DeepSWE v1.1
long-horizon engineering		 69.6%72.7%67%69.7%
Terminal-Bench 2.1
terminal work		 87.4%88.8%85.6%83.1%
BrowseComp
browsing		 87.5%90.4%84.4%not published
OSWorld 2.0
computer use		 50.2%62.6%47.5%not published
MRCR v2, 256K to 512K
long context		 89.6%91.5%81.5%not published
FrontierMath Tier 4
hard math		 68.3%83%72.5%87.8%
Toolathlon
tool use		 53.1%58%55.6%61.7%

Table from the GPT-5.6 announcement, Jul 9, 2026. The dot marks the best in each row. "Not published" cells are tests where OpenAI did not show Fable 5.

Terra between Sol and Luna

Within the family, Terra costs half of Sol for input and 60% for output. Luna costs a tenth of Terra and comes close to it on several tests. The gap shows up in long context and computer use, where Terra is well ahead.

The GPT-5.6 family, price per million tokens and launch scores
GPT-5.6 SolGPT-5.6 TerraGPT-5.6 Luna
Input and output today US$ 4 and US$ 20US$ 2 and US$ 12US$ 0.20 and US$ 1.20
Cache read US$ 0.40US$ 0.20US$ 0.02
Launch price US$ 5 and US$ 30US$ 2.50 and US$ 15US$ 1 and US$ 6
Equivalent old tier unsuffixedmininano
Agents' Last Exam 52.7%50.4%50.3%
DeepSWE v1.1 72.7%69.6%67.2%
OSWorld 2.0 62.6%50.2%45.6%
MRCR v2, 256K to 512K 91.5%89.6%41.3%

OpenAI pricing page (Oct 1, 2026), model pages and the GPT-5.6 announcement table. Sol price is the promotional one. The dot marks the best in each score row.

Is GPT-5.6 Terra still worth using in 2026?

For a new project, it is hard to justify. OpenAI did not release a Terra in generation 6, and GPT-6.1 Sol, the current middle tier, costs the same for input, less for output and half for cache. GPT-6 Sol also ties or comes in below Terra on every line of the price table.

OpenAI did not include Terra in the GPT-6 family performance charts. The direct official comparisons sit in the safety reports: on HealthBench Professional, Terra scores 57.7, against 64.2 for 6.1 Sol; on agents, it has 13.5% misaligned outcomes, against 3.4% for Astra.

For teams already using it, there is no rush: the deprecations page shows no date for Terra and still lists it as the replacement for gpt-5-mini and o4-mini. OpenAI's reasoning guide also still mentions it as a cheaper option.

GPT-5.6 Terra against the current lineup, price per million tokens
GPT-5.6 TerraGPT-6.1 SolGPT-6 SolGPT-6 Luna
Input US$ 2US$ 2US$ 2US$ 0.10
Output US$ 12US$ 10US$ 10US$ 0.50
Cache read US$ 0.20US$ 0.10US$ 0.20US$ 0.01
Cache write US$ 2.50US$ 2.50US$ 2.50US$ 0.125
Above 272 thousand tokens
input and output		 US$ 4 and US$ 18US$ 4 and US$ 15US$ 4 and US$ 15US$ 0.20 and US$ 0.75
Batch and Flex
input and output		 US$ 1 and US$ 6US$ 1 and US$ 5US$ 1 and US$ 5US$ 0.05 and US$ 0.25
Knowledge cutoff Feb 16, 2026Apr 30, 2026Apr 20, 2026May 18, 2026
Accepts effort none yesnoyesyes
HealthBench Professional
length-adjusted		 57.764.260.860.8

OpenAI documentation, pricing page and model pages, checked on Oct 1, 2026. HealthBench from table 7 of the GPT-6.1 Sol safety report; the dot marks the best in that row only.

What changes when swapping Terra for a GPT-6 model

OpenAI's migration guide lists points that break integrations:

  • GPT-6.1 Sol does not accept none effort (OpenAI says to use low) and only calls tools through the Responses API.
  • GPT-6 Sol keeps none, but on Chat Completions it only calls functions at that effort.
  • With effort other than none, temperature, top_p and top_logprobs go away.
  • Reasoning saved in the 5.6 family does not carry over to 6, because OpenAI only allows reuse within the same family.
  • 6.1 Sol is rated Critical in cybersecurity and uses GPT-6 Astra's safeguards; Terra is High, with the GPT-5.6 package.
In practice

Examples of using GPT-5.6 Terra

Reading long contracts

A whole set of contracts goes into a single request. Terra scored 89.6% on OpenAI's needle test between 256 thousand and 512 thousand tokens; above 272 thousand input tokens, the whole request moves to US$ 4 input and US$ 18 output.

Leaving gpt-5-mini before shutdown

gpt-5-mini-2025-08-07 shuts down on Dec 11, 2026, and the official table points to Terra as the replacement. It is worth running the same tasks on GPT-6.1 Sol, which costs the same for input and less for output, and choosing by the result.

An office agent with confirmations

An agent that sends emails and changes company systems asks for approval before sensitive actions. In OpenAI's test, the confirmation policy cut Terra's misaligned outcomes from 13.5% to 5.7%.

What the tests say

Where GPT-5.6 Terra stands out and where it falls behind, by the numbers

GPT-5.6 Terra has published results on 28 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Oct 12, 2025 and Oct 7, 2026 that took the same test. Against the whole archive, which includes 2023 models, a recent release lands near the top of almost everything, and that says nothing.

Where it stands out

Tests where it sits in the top third of its own generation.

CritPt30.0%
#9 of 79 in its generation, tied with 1 · #9 of 112 overall
ALE-Bench1,951
#8 of 65 in its generation · #8 of 82 overall
Lmca54.95
#11 of 70 in its generation · #11 of 121 overall
Frontiermath Tiers 1 3 V286.0%
#10 of 66 in its generation, tied with 1 · #10 of 86 overall
AIME 2024 and 2025 (OTIS)99.7%
#13 of 82 in its generation · #13 of 168 overall

Where it falls behind

Tests where it sits in the bottom half of its own generation.

SimpleBench48.9%
#30 of 38 in its generation · record: 81.9% (Claude Fable 5)
CursorBench41.3%
#12 of 16 in its generation · record: 57.8% (Claude Opus 5.5)
SimpleQA Verified43.2%
#37 of 64 in its generation · record: 76.0% (Gemini 4 Argon)
BALROG53.2%
#6 of 10 in its generation · record: 68.3% (GPT-6 Astra)
FrontierCode41.3%
#23 of 43 in its generation · record: 54.6% (Claude Opus 5.5)
See all 28 tests, one by one
TestScoreIn its generationOverallTest record
GPQA Diamond 93.3% #17 / 81 #17 / 173 96.4% Gemini 4 Argon
AIME 2024 and 2025 (OTIS) 99.7% #13 / 82 #13 / 168 100.0% GPT-5.5
Dtbench 93.3% #26 / 70 #26 / 148 98.9% Claude Opus 5.5
Lmca 54.95 #11 / 70 #11 / 121 68.23 Claude Opus 5.5
CritPt 30.0% #9 / 79 (tied with 1) #9 / 112 32.3% GPT-5.6 Sol
WeirdML 78.3% #11 / 55 #11 / 112 93.6% GPT-6 Astra
SciCode 55.0% #25 / 76 (tied with 2) #26 / 104 66.9% Claude Opus 5.5
Frontiermath Tiers 1 3 V2 86.0% #10 / 66 (tied with 1) #10 / 86 93.7% GPT-6 Astra
ARC-AGI 2 83.9% #13 / 51 #13 / 83 95.0% GPT-6 Astra
ALE-Bench 1,951 #8 / 65 #8 / 82 2,951 GPT-6 Astra
WebDev Arena 1,521 #35 / 76 #35 / 82 1,820 Claude Opus 5.5
ARC-AGI 96.5% #10 / 51 (tied with 1) #10 / 81 98.5% Claude Fable 5
SimpleQA Verified 43.2% #37 / 64 #41 / 81 76.0% Gemini 4 Argon
Chess puzzles 47.0% #21 / 65 (tied with 1) #22 / 80 76.0% GPT-6 Astra
Game puzzles 35.0% #19 / 61 (tied with 1) #20 / 77 84.0% GPT-6 Astra
ProofBench 74.0% #14 / 69 #15 / 74 100.0% Claude Fable 5.1
SimpleBench 48.9% #30 / 38 #38 / 74 81.9% Claude Fable 5
Frontiermath Tier 4 V2 70.7% #13 / 56 #13 / 67 100.0% GPT-6.1 Sol
Vending Bench 2 7,343 #14 / 49 #15 / 57 15,515 GPT-6 Astra
Apex Agents 58.2% #14 / 49 #14 / 50 82.4% Gemini 4 Argon
FrontierCode 41.3% #23 / 43 #23 / 43 54.6% Claude Opus 5.5
Gdp Pdf 24.7% #12 / 36 #13 / 37 35.0% GPT-6.1 Sol
Furniture Assembly 54.2% #10 / 33 #10 / 34 83.3% Claude Opus 5.5
BALROG 53.2% #6 / 10 #6 / 32 68.3% GPT-6 Astra
Blueprint Bench 2 30.8% #16 / 31 #16 / 31 54.4% Gemini 4 Argon
DeepSWE 69.6% #6 / 26 #6 / 26 74.1% GPT-6 Astra
Surface Evolver Bench 83.8% #6 / 23 #6 / 24 95.0% Claude Fable 5
CursorBench 41.3% #12 / 16 #12 / 16 57.8% Claude Opus 5.5

For each test, the model's best effort level counts, as in the ranking. Source of the scores: Epoch AI, under CC BY 4.0.

Comparison

What to compare GPT-5.6 Terra with

The models from other labs closest in capability and released in the same period, at most two per lab. Click any of them to see its page.

ModelLabECIELOLiveBenchRelease
GPT-5.6 Terra OpenAI 159.43 1,515 77.9 Jul 9, 2026
Claude Haiku 5.5 Anthropic 158.98 n/a 72.1 Oct 7, 2026
Claude Opus 4.8 Anthropic 158.26 1,507 76.2 May 28, 2026
Kimi K3 Moonshot 157.42 1,528 79.2 Jul 16, 2026
Gemini 3.7 Flash Google DeepMind 157.22 1,551 78.8 Aug 13, 2026
Gemini 3.8 Flash Google DeepMind 157.20 1,551 75.8 Sep 2, 2026
Qwen 3.8 Max Alibaba 156.88 1,536 78.5 Aug 2, 2026
Head to head

GPT-5.6 Terra or Claude Haiku 5.5?

Claude Haiku 5.5 is the model from another lab closest to GPT-5.6 Terra in capability. Across the 16 tests both took, this is the score:

GPT-5.6 Terra ahead on 10 Claude Haiku 5.5 ahead on 5 tied on 1
Largest differences first
TestGPT-5.6 TerraClaude Haiku 5.5
Vending Bench 2 7,343 3,602
SimpleQA Verified 43.2% 23.8%
Frontiermath Tier 4 V2 70.7% 46.3%
CritPt 30.0% 22.6%
ProofBench 74.0% 89.0%
CursorBench 41.3% 48.4%
Chess puzzles 47.0% 55.0%
Game puzzles 35.0% 30.0%
Frontiermath Tiers 1 3 V2 86.0% 75.1%
Furniture Assembly 54.2% 47.5%

Unweighted score: each test counts one point, whether math or coding. The dot marks who did better on each row.

Cost per task

What GPT-5.6 Terra costs to deliver each task

Price per token is what is on the price sheet; cost per task is what lands on the invoice, because it depends on how many tokens the model spends. Each point is an effort level, published by whoever ran the test and compiled by Epoch AI. The further up and to the left, the more score per dollar.

ARC-AGI 2

The score scale does not start at zero.

2%25%48%71%95%US$ 0.20US$ 0.50US$ 1US$ 2cost per task (log scale)GPT-5.6 Terra · low · 18.8% · US$ 0.17lowGPT-5.6 Terra · medium · 37.5% · US$ 0.27mediumGPT-5.6 Terra · high · 67.1% · US$ 0.55highGPT-5.6 Terra · xhigh · 74.2% · US$ 0.69xhighGPT-5.6 Terra · max · 83.9% · US$ 1.09maxClaude Opus 4.8 · low · 62.2% · US$ 1.68Claude Opus 4.8 · medium · 71.7% · US$ 2.39Claude Opus 4.8 · high · 72.1% · US$ 2.74Kimi K3 · low · 12.4% · US$ 0.25Kimi K3 · high · 55.0% · US$ 0.95Kimi K3 · max · 60.4% · US$ 1.59
GPT-5.6 TerraClaude Opus 4.8Kimi K3

ARC-AGI

The score scale does not start at zero.

55%66%77%89%100%US$ 0.10US$ 0.20US$ 0.50US$ 1US$ 2cost per task (log scale)GPT-5.6 Terra · low · 60.2% · US$ 0.09lowGPT-5.6 Terra · medium · 77.0% · US$ 0.13mediumGPT-5.6 Terra · high · 92.0% · US$ 0.19highGPT-5.6 Terra · xhigh · 94.0% · US$ 0.26xhighGPT-5.6 Terra · max · 96.5% · US$ 0.55maxClaude Opus 4.8 · low · 88.0% · US$ 0.67Claude Opus 4.8 · medium · 91.5% · US$ 0.91Claude Opus 4.8 · high · 92.0% · US$ 1.04Claude Opus 4.8 · max · 92.5% · US$ 2.33Kimi K3 · low · 65.7% · US$ 0.18Kimi K3 · high · 86.7% · US$ 0.48Kimi K3 · max · 94.5% · US$ 0.77
GPT-5.6 TerraClaude Opus 4.8Kimi K3

CursorBench

The score scale does not start at zero.

22%29%37%44%52%US$ 0.10US$ 0.20US$ 0.50US$ 1US$ 2US$ 5cost per task (log scale)GPT-5.6 Terra · low · 25.2% · US$ 0.52lowGPT-5.6 Terra · medium · 27.6% · US$ 0.64mediumGPT-5.6 Terra · high · 30.7% · US$ 1.11highGPT-5.6 Terra · xhigh · 33.6% · US$ 1.81xhighGPT-5.6 Terra · max · 41.3% · US$ 5.14maxClaude Haiku 5.5 · low · 30.9% · US$ 0.08Claude Haiku 5.5 · medium · 36.9% · US$ 0.17Claude Haiku 5.5 · high · 42.3% · US$ 0.32Claude Haiku 5.5 · xhigh · 44.3% · US$ 0.56Claude Haiku 5.5 · max · 48.4% · US$ 1.12Gemini 3.8 Flash · medium · 37.3% · US$ 4.06Gemini 3.8 Flash · high · 39.6% · US$ 4.70
GPT-5.6 TerraClaude Haiku 5.5Gemini 3.8 Flash
Family

Other models from OpenAI

The models from the same lab closest in time, newest first.

ModelReleaseECIELOLiveBenchAccess
GPT-6 Sol Sep 22, 2026 162.55 n/a 79.3 API access
GPT-6 Luna Sep 22, 2026 156.31 n/a 72.0 API access
GPT-6 Astra Sep 3, 2026 166.41 n/a 82.2 API access
GPT-5.6 Sol Jul 9, 2026 161.73 1,527 81.1 API access
GPT-5.6 Terra Jul 9, 2026 159.43 1,515 77.9 API access
GPT-5.6 Luna Jul 9, 2026 156.33 1,478 73.6 API access
GPT-5.5 Instant May 5, 2026 142.45 1,456 n/a API access
GPT-5.5 Pro Apr 23, 2026 161.97 n/a n/a API access
GPT-5.5 Apr 23, 2026 158.95 1,533 80.2 API access
Your rating

Have you used GPT-5.6 Terra? Rate it

n/a out of 5
no votes yet

Users' opinion, not measurement: the rating is kept apart from the ranking and does not move anyone's position. One vote per browser, no sign-up, and you can change your rating later.

Tap a star to vote.
Frequently asked questions

What people ask about GPT-5.6 Terra

How much does GPT-5.6 Terra cost?

US$ 2 per million input tokens and US$ 12 per million output, since the 20% cut on Jul 30, 2026; at launch it was US$ 2.50 and US$ 15. Cache reads cost US$ 0.20 and cache writes US$ 2.50. Above 272 thousand input tokens, the whole request moves to US$ 4 and US$ 18. Batch and Flex are half price, and Fast mode costs double.

What is the difference between GPT-5.6 Terra, Sol and Luna?

They are the three tiers of the GPT-5.6 family. Sol is the flagship, Terra is the balanced lower-cost model and Luna is the fastest and cheapest. OpenAI compares Sol to the old unsuffixed model, Terra to mini and Luna to nano.

Is there a GPT-6 Terra?

No. As of Oct 1, 2026, the GPT-6 family has GPT-6 Astra, GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna, and no official OpenAI page mentions a GPT-6 Terra. In the GPT-5.6 announcement, OpenAI said each tier advances on its own cadence.

Will GPT-5.6 Terra be retired?

No date has been published. OpenAI's deprecations page does not list Terra and still names it as the replacement for models that shut down on Oct 23, 2026 and Dec 11, 2026. OpenAI promises at least 6 months' notice for generally available models.

Can I use Terra in ChatGPT?

Not in regular Chat conversations. It appears in ChatGPT Work and Codex, depending on the plan, and in Codex it is also available on the Free and Go plans. Developers use it through the OpenAI API.

Where does GPT-5.6 Terra rank among AI models?

In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, GPT-5.6 Terra ranks #12 out of 267 models, with an ECI of 159.43. The ECI is a composite index that combines several hard benchmarks into a single scale. The position changes when a new model comes out, and the page is refreshed six times a day.

Is GPT-5.6 Terra better than Claude Haiku 5.5?

Across the 16 tests both models took and Epoch AI compiled, GPT-5.6 Terra comes out ahead on 10, behind on 5 and tied on 1. It is a test-by-test count with no weighting: a win in math counts the same as a win in coding. To choose between the two, look at the tests that resemble your own use.

What is GPT-5.6 Terra best at?

Against models of the same generation (released up to 9 months before and up to 3 months after it), the best relative results of GPT-5.6 Terra are: CritPt (#9 of 79); ALE-Bench (#8 of 65); Lmca (#11 of 70). The yardstick is its position among the models of that period that took the same test.

Where do the numbers on this page come from?

From the same three sources as the ranking, all openly licensed: Epoch AI (ECI and individual tests, CC BY 4.0), Arena (human-preference ELO, CC BY 4.0) and LiveBench (overall average, CC BY-SA 4.0). ROO3 does not measure any model. When the page shows a figure published by the maker itself, it appears in the written analysis, labeled as the maker's, and never enters the ranking.

Sources

Where each number comes from

  1. OpenAI, GPT-5.6 announcement
  2. OpenAI, GPT-5.6 Terra model page (documentation)
  3. OpenAI, API pricing
  4. OpenAI, API changelog (launch and Jul 30, 2026 price cut)
  5. OpenAI, GPT-5.6 system card
  6. OpenAI, GPT-6 Astra system card (table 8, including GPT-5.6 Terra)
  7. OpenAI, GPT-6.1 Sol safety report (table 7, including GPT-5.6 Terra)
  8. OpenAI, API deprecations
  9. OpenAI, latest model guide (migration)
  10. OpenAI, reasoning guide
  11. OpenAI, models on Amazon Bedrock (documentation)
  12. OpenAI Help Center, GPT-5.6 and GPT-6 Pro in ChatGPT

Epoch AI, "AI Benchmarking Hub". Published at epoch.ai. Data under the CC BY 4.0 licence, reordered by roo3.co.

The Arena (LMArena) leaderboard, from the public lmarena-ai/leaderboard-dataset on Hugging Face, under the CC BY 4.0 licence. We are not affiliated with the Arena.

LiveBench, by the LiveBench team, published at livebench.ai under the CC BY-SA 4.0 license. The overall average is recomputed by roo3.co with the same formula as the website (average of the categories), and those numbers stay under CC BY-SA 4.0. We are not affiliated with LiveBench.

← See the full ranking