Home › AI Benchmark › GPT-5.5
OpenAI released on Apr 23, 2026 API access

GPT-5.5

GPT-5.5 was OpenAI's main model from Apr 23, 2026 until GPT-5.6 arrived in July. It costs US$ 5 per million input tokens and US$ 30 per million output, 2.5 times the input and 3 times the output price of GPT-6.1 Sol, the current mid-tier model. It leaves ChatGPT, ChatGPT Work and Codex on Oct 14, 2026, but stays on the API with no published shutdown date. Keep it where it is already tuned; new projects should start on the GPT-6 family.

#15 of 267 in the ECI ranking
158.95 ECI, the Epoch AI capability index
1,533 Arena ELO, human preference · #14
80.2 LiveBench, overall average from 0 to 100 · #9
5.0 out of 5 · 1 vote
Spec sheet

GPT-5.5 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-5.5 (snapshot gpt-5.5-2026-04-23)
Release
Apr 23, 2026 in ChatGPT and Codex; Apr 24, 2026 on the API
Input
US$ 5 per million tokens
Output
US$ 30 per million tokens
Cache read
US$ 0.50 per million (10% of input), no charge for cache writes
Above 272 thousand input tokens
input doubled and output 1.5x, for the whole session: US$ 10, US$ 1 for cache and US$ 45
Batch and Flex
half price: US$ 2.50 input and US$ 15 output
Fast mode
US$ 12.50 and US$ 75, 2.5 times the standard rate; called Priority until Jul 30, 2026
Context window
1,050,000 tokens on the API; 400 thousand in Codex
Max output
128 thousand tokens
Knowledge cutoff
Dec 1, 2025
Reasoning
effort none, low, medium (default), high and xhigh
Where it runs
OpenAI API (Responses, Chat Completions and Batch) and Amazon Bedrock; ChatGPT, Work and Codex until Oct 14, 2026
Cyber risk
High on OpenAI's Preparedness scale, below Critical
Status
leaves ChatGPT, Work and Codex on Oct 14, 2026; no deprecation notice on the API
Analysis

Where GPT-5.5 does well, where it falls behind and who it is for

Where it does well

  • Coding in the terminal: 82.7% on Terminal-Bench 2.0 in OpenAI's table, against 75.1% for GPT-5.4, 69.4% for Claude Opus 4.7 and 68.5% for Gemini 3.1 Pro.
  • Long context: 74.0% on MRCR v2, which looks for eight pieces of information between 512 thousand and 1 million tokens, against 36.6% for GPT-5.4 and 32.2% for Opus 4.7.
  • Office work: 84.9% on GDPval, the top score in the table, above even GPT-5.5 Pro (82.3%).
  • Computer use: 78.7% on OSWorld-Verified, ahead of Opus 4.7 (78.0%) and GPT-5.4 (75.0%).
  • Still answers without reasoning: none effort works on GPT-5.5, while GPT-6 Astra and GPT-6.1 Sol refuse it.

Where it falls behind

  • Price against the current line: US$ 5 and US$ 30 per million tokens, 2.5 times the input and 3 times the output of GPT-6.1 Sol, which OpenAI describes as near-Astra performance at a lower cost.
  • Coding in real repositories: 58.6% on SWE-Bench Pro, against 64.3% for Claude Opus 4.7 in OpenAI's own table, which notes evidence of memorization on this test.
  • Hard knowledge: 41.4% on Humanity's Last Exam without tools, behind Opus 4.7 (46.9%) and Gemini 3.1 Pro (44.4%).
  • A more eager agent: in the safety report it came out slightly more misaligned than GPT-5.4 Thinking on coding work, such as acting when the user was only asking. Nearly all of it was low severity, according to OpenAI.

Who it makes sense for

  • Production systems already tuned and measured on GPT-5.5 through the API, while you test the switch: there it has no published shutdown date.
  • Data residency in the United Arab Emirates: OpenAI lists GPT-5.5 in that region, which needs extra approval, and no GPT-6 model appears there.
  • Evaluations that must be reproducible: the gpt-5.5-2026-04-23 snapshot stays fixed and works as a baseline to measure what a new model gains.
  • Anyone using GPT-5.5 in Codex or ChatGPT: the time until Oct 14, 2026 is for replacing gpt-5.5 in workspace defaults, agents, scheduled tasks and scripts.

When to pick another model

  • New projects: GPT-6.1 Sol costs US$ 2 and US$ 10 and is the cost-and-intelligence balance option in OpenAI's current guide.
  • High volume of simple tasks, such as classifying and extracting: GPT-6 Luna costs US$ 0.10 and US$ 0.50, 50 times less on input and 60 times less on output.
  • Frequent requests above 272 thousand tokens: the whole session moves to US$ 10 input and US$ 45 output.
  • Work in ChatGPT or Codex after Oct 14, 2026: in Codex, OpenAI points to GPT-6 Sol on paid plans and GPT-6 Luna on Free and Go.

What OpenAI published at launch

GPT-5.5 reached ChatGPT and Codex on Apr 23, 2026 and the API the next day. OpenAI said it matched GPT-5.4 per-token latency and used significantly fewer tokens on the same Codex tasks, and used that efficiency to justify doubling the price per token: from US$ 2.50 and US$ 15 to US$ 5 and US$ 30.

The table below is the one from the announcement, with GPT-5.5 at xhigh effort in a research environment. OpenAI picked the tests and the rivals, and even so GPT-5.5 loses four of these rows: SWE-Bench Pro to Claude Opus 4.7, BrowseComp to Gemini 3.1 Pro, and MCP Atlas and Humanity's Last Exam to both.

GPT-5.5 in the figures OpenAI published
GPT-5.5
OpenAI		GPT-5.4
OpenAI		Claude Opus 4.7
Anthropic		Gemini 3.1 Pro
Google
Terminal-Bench 2.0
coding in the terminal		 82.7%75.1%69.4%68.5%
SWE-Bench Pro
fixing real repository issues		 58.6%57.7%64.3%54.2%
GDPval
office work		 84.9%83.0%80.3%67.3%
OSWorld-Verified
computer use		 78.7%75.0%78.0%not measured
BrowseComp
web research		 84.4%82.7%79.3%85.9%
MCP Atlas
tool use		 75.3%70.6%79.1%78.2%
FrontierMath Tier 4
research math		 35.4%27.1%22.9%16.7%
Humanity's Last Exam
hard knowledge, no tools		 41.4%39.8%46.9%44.4%
CyberGym
cybersecurity		 81.8%79.0%73.1%not measured
MRCR v2, 512 thousand to 1 million
long context		 74.0%36.6%32.2%not measured
ARC-AGI-2
abstract reasoning		 85.0%73.3%75.8%77.1%

Figures from the GPT-5.5 announcement (Apr 23, 2026), at xhigh effort. The dot marks the best in each row. "Not measured" means OpenAI published no figure for that model. On SWE-Bench Pro, OpenAI notes evidence of memorization.

Is GPT-5.5 still worth using in 2026?

Two generations came after it. GPT-5.6 came out on Jul 9, and OpenAI described Terra, the middle tier, as a lower-cost model with performance competitive with GPT-5.5. September brought GPT-6 Astra, GPT-6 Sol, GPT-6 Luna and GPT-6.1 Sol. OpenAI's current model guide names only Astra, 6.1 Sol and Luna.

The date that matters is Oct 14, 2026, when it leaves ChatGPT, ChatGPT Work and Codex on every plan. The retirement does not apply to the API, and the deprecations page has no entry for gpt-5.5: API users do not need to rush, but they pay more than on the current line.

Per token, it is only cheaper than Astra: half on input and 60% on output. The reasons to stay are different: prompts and evals already tuned, data residency in the Emirates, or a fixed snapshot. Otherwise the price already favors switching, and a test on your own tasks tells you whether quality keeps up.

GPT-5.5 and OpenAI's current line, price per million tokens
GPT-5.5
Apr 2026		GPT-6 Astra
Sep 2026		GPT-6.1 Sol
Sep 2026		GPT-6 Luna
Sep 2026
Input US$ 5US$ 10US$ 2US$ 0.10
Output US$ 30US$ 50US$ 10US$ 0.50
Cache read US$ 0.50US$ 1US$ 0.10US$ 0.01
Cache write no chargeUS$ 12.50US$ 2.50US$ 0.125
Above 272 thousand tokens
input and output		 US$ 10 and US$ 45US$ 20 and US$ 75US$ 4 and US$ 15US$ 0.20 and US$ 0.75
Knowledge cutoff Dec 1, 2025Apr 30, 2026Apr 30, 2026May 18, 2026
Accepts effort none yesnonoyes
Status leaves ChatGPT on Oct 14, 2026currentcurrentcurrent

OpenAI documentation, pricing page and model pages, checked on Oct 1, 2026. Separately billed cache writes started with GPT-5.6.

What changes when moving from GPT-5.5 to GPT-6

At launch, OpenAI asked developers to tune GPT-5.5 from scratch, not treat it as a drop-in replacement for GPT-5.4. The same care applies on the way out:

  • GPT-6 Astra and GPT-6.1 Sol do not accept none effort; OpenAI says to use low instead. GPT-6 Sol and GPT-6 Luna accept it.
  • On Astra and 6.1 Sol, tool calls only go through the Responses API. Chat Completions answers but does not call functions.
  • With any effort other than none, temperature, top_p and top_logprobs must go.
  • Caching changes: replace prompt_cache_retention with prompt_cache_options.ttl set to "30m". Writes now cost 1.25 times input, and reads 10% of it (5% on GPT-6.1 Sol).
  • Astra stops to ask more often and is more sensitive to instructions in files such as AGENTS.md. OpenAI recommends auditing those files.
In practice

Examples of using GPT-5.5

Customer support with tools

An assistant looks up orders, applies company policy and resolves the case end to end. On Tau2-bench Telecom, which measures this kind of flow, GPT-5.5 scored 98.0% without prompt tuning, in OpenAI's figures.

Reading an entire repository or archive

Code and documentation go in together, and the model cross-checks scattered information. GPT-5.5 scored 74.0% on MRCR v2 between 512 thousand and 1 million tokens, but above 272 thousand tokens the whole session moves to US$ 10 and US$ 45.

A side-by-side migration test

Run ten real tasks on GPT-5.5 and GPT-6.1 Sol with the same prompt and compare accuracy and cost per task. The fixed GPT-5.5 snapshot keeps the baseline stable.

What the tests say

Where GPT-5.5 stands out and where it falls behind, by the numbers

GPT-5.5 has published results on 43 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Jul 27, 2025 and Jul 22, 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.

FrontierMath51.7%
#1 of 37 in its generation · #1 of 65 overall
Terminal-Bench84.7%
#1 of 31 in its generation · #1 of 33 overall
Cl Bench Life22.2%
#1 of 13 in its generation · #1 of 13 overall
Exploitbench47.4%
#1 of 8 in its generation · #1 of 8 overall
AIME 2024 and 2025 (OTIS)100.0%
#1 of 71 in its generation, tied with 3 · #1 of 168 overall

Where it falls behind

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

MirrorCode10.0%
#5 of 6 in its generation · record: 77.4% (Claude Opus 5.5)
Posttrainbench27.2%
#7 of 10 in its generation · record: 41.8% (Claude Fable 5)
See all 43 tests, one by one
TestScoreIn its generationOverallTest record
GPQA Diamond 94.0% #4 / 71 #11 / 173 96.4% Gemini 4 Argon
AIME 2024 and 2025 (OTIS) 100.0% #1 / 71 (tied with 3) #1 / 168 100.0% GPT-5.5
Dtbench 96.0% #4 / 67 (tied with 2) #12 / 148 98.9% Claude Opus 5.5
Lmca 54.27 #6 / 65 #12 / 121 68.23 Claude Opus 5.5
CritPt 27.1% #6 / 60 #14 / 112 32.3% GPT-5.6 Sol
WeirdML 84.9% #3 / 59 #6 / 112 93.6% GPT-6 Astra
SciCode 56.1% #7 / 56 #24 / 104 66.9% Claude Opus 5.5
Frontiermath Tiers 1 3 V2 85.3% #5 / 50 #13 / 86 93.7% GPT-6 Astra
ARC-AGI 2 85.0% #3 / 42 #10 / 83 95.0% GPT-6 Astra
ALE-Bench 1,943 #4 / 58 #9 / 82 2,951 GPT-6 Astra
WebDev Arena 1,510 #16 / 58 #39 / 82 1,820 Claude Opus 5.5
ARC-AGI 95.0% #6 / 42 #14 / 81 98.5% Claude Fable 5
SimpleQA Verified 63.0% #7 / 48 #14 / 81 76.0% Gemini 4 Argon
Chess puzzles 63.0% #2 / 47 #6 / 80 76.0% GPT-6 Astra
Game puzzles 56.0% #2 / 45 #9 / 77 84.0% GPT-6 Astra
ProofBench 50.0% #10 / 45 (tied with 1) #29 / 74 100.0% Claude Fable 5.1
SimpleBench 69.0% #10 / 41 #13 / 74 81.9% Claude Fable 5
Frontiermath Tier 4 V2 72.5% #4 / 40 #12 / 67 100.0% GPT-6.1 Sol
FrontierMath 51.7% #1 / 37 #1 / 65 51.7% GPT-5.5
ForecastBench 60.80 #12 / 30 #19 / 62 62.50 o3
Vending Bench 2 7,524 #5 / 43 #14 / 57 15,515 GPT-6 Astra
Apex Agents 55.1% #4 / 27 #19 / 50 82.4% Gemini 4 Argon
FrontierMath tier 4 35.4% #2 / 37 #2 / 48 39.6% GPT-5.5 Pro
FrontierCode 43.0% #5 / 22 #19 / 43 54.6% Claude Opus 5.5
Gdp Pdf 26.0% #3 / 18 #12 / 37 35.0% GPT-6.1 Sol
Furniture Assembly 44.2% #3 / 16 #13 / 34 83.3% Claude Opus 5.5
Terminal-Bench 84.7% #1 / 31 #1 / 33 84.7% GPT-5.5
SWE-bench Verified 80.6% #2 / 26 #2 / 32 83.5% Claude Opus 4.7
Blueprint Bench 2 36.2% #2 / 20 #8 / 31 54.4% Gemini 4 Argon
Gso 40.2% #6 / 17 #8 / 28 88.2% Claude Fable 5.1
Ebr Bench 34.3% #3 / 17 #11 / 27 76.2% GPT-6 Astra
DeepSWE 67.0% #6 / 17 #11 / 26 74.1% GPT-6 Astra
DeepResearch Bench 54.0% #4 / 17 #4 / 24 55.3% Claude Opus 4.6
Surface Evolver Bench 88.1% #4 / 19 #4 / 24 95.0% Claude Fable 5
Gbaeval 53.2% #5 / 21 #6 / 22 79.6% Claude Opus 5
Rli 6.3% #3 / 9 #6 / 15 21.3% Claude Opus 5.5
Cl Bench Life 22.2% #1 / 13 #1 / 13 22.2% GPT-5.5
Posttrainbench 27.2% #7 / 10 #8 / 11 41.8% Claude Fable 5
MirrorCode 10.0% #5 / 6 #8 / 9 77.4% Claude Opus 5.5
Osworld 2 13.0% #4 / 8 #5 / 9 31.4% Claude Opus 5
Btf3 14.3% #1 / 5 #3 / 8 14.9% GLM-5.3
Exploitbench 47.4% #1 / 8 #1 / 8 47.4% GPT-5.5
Frontiermath Erdos 0.0% #1 / 3 (tied with 2) #4 / 7 2.9% GPT-6 Astra

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-5.5 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
Claude Haiku 5.5 Anthropic 158.98 n/a 72.1 Oct 7, 2026
GPT-5.5 OpenAI 158.95 1,533 80.2 Apr 23, 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-5.5 or Claude Haiku 5.5?

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

GPT-5.5 ahead on 11 Claude Haiku 5.5 ahead on 5
Largest differences first
TestGPT-5.5Claude Haiku 5.5
SimpleQA Verified 63.0% 23.8%
Vending Bench 2 7,524 3,602
Game puzzles 56.0% 30.0%
ProofBench 50.0% 89.0%
Frontiermath Tier 4 V2 72.5% 46.3%
Ebr Bench 34.3% 43.8%
CritPt 27.1% 22.6%
Chess puzzles 63.0% 55.0%
Frontiermath Tiers 1 3 V2 85.3% 75.1%
Blueprint Bench 2 36.2% 33.0%

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-5.5 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.

1%25%49%72%96%US$ 0.50US$ 1US$ 2cost per task (log scale)GPT-5.5 · low · 33.3% · US$ 0.35lowGPT-5.5 · medium · 70.4% · US$ 0.86mediumGPT-5.5 · high · 83.3% · US$ 1.45highGPT-5.5 · xhigh · 85.0% · US$ 1.87xhighClaude 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-5.5Claude Opus 4.8Kimi K3

ARC-AGI

The score scale does not start at zero.

61%71%80%90%99%US$ 0.20US$ 0.50US$ 1US$ 2cost per task (log scale)GPT-5.5 · low · 76.2% · US$ 0.20lowGPT-5.5 · medium · 92.2% · US$ 0.39mediumGPT-5.5 · high · 94.5% · US$ 0.56highGPT-5.5 · xhigh · 95.0% · US$ 0.73xhighClaude 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-5.5Claude Opus 4.8Kimi K3

DeepSWE

The score scale does not start at zero.

20%35%50%66%81%US$ 2US$ 5US$ 10cost per task (log scale)GPT-5.5 · low · 27.0% · US$ 1.20lowGPT-5.5 · medium · 54.0% · US$ 2.75mediumGPT-5.5 · high · 64.4% · US$ 5.10highGPT-5.5 · xhigh · 67.0% · US$ 7.23xhighClaude Opus 4.8 · low · 40.8% · US$ 2.29Claude Opus 4.8 · medium · 48.7% · US$ 3.44Claude Opus 4.8 · high · 51.8% · US$ 4.28Claude Opus 4.8 · xhigh · 54.4% · US$ 8.01Claude Opus 4.8 · max · 59.0% · US$ 13.22Gemini 3.7 Flash · low · 53.8% · US$ 1.83Gemini 3.7 Flash · medium · 65.5% · US$ 2.03Gemini 3.7 Flash · high · 65.3% · US$ 2.18Gemini 3.8 Flash · medium · 71.0% · US$ 1.97Gemini 3.8 Flash · high · 73.8% · US$ 2.36
GPT-5.5Claude Opus 4.8Gemini 3.7 FlashGemini 3.8 Flash
Family

Other models from OpenAI

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

ModelReleaseECIELOLiveBenchAccess
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-5.5 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
GPT-5.4 Mini Mar 17, 2026 148.77 1,446 66.4 API access
GPT-5.4 Nano Mar 17, 2026 146.12 1,404 69.6 API access
GPT-5.4 Pro Mar 5, 2026 159.00 n/a n/a API access
GPT-5.4 Mar 5, 2026 156.72 1,519 78.0 API access
Your rating

Have you used GPT-5.5? Rate it

5.0 out of 5
1 vote

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-5.5

How much does GPT-5.5 cost?

US$ 5 per million input tokens and US$ 30 per million output, with cache reads at US$ 0.50 and no charge for cache writes. Above 272 thousand input tokens, the whole session moves to US$ 10 and US$ 45. Batch and Flex are half price, and the price has not changed since launch.

Is GPT-5.5 being shut down?

In ChatGPT, ChatGPT Work and Codex, yes, on Oct 14, 2026, on every plan. On the API there is no date: OpenAI says the retirement does not apply there, and the deprecations page does not list the model.

Which model replaces GPT-5.5?

In Codex, OpenAI points to GPT-6 Sol for the Plus, Pro, Business, Enterprise and Edu plans, and to GPT-6 Luna for Free and Go. On the API, the current guide has three options: GPT-6 Astra for the most demanding work, GPT-6.1 Sol as the balance of cost and intelligence, and GPT-6 Luna for volume. At US$ 2 and US$ 10, 6.1 Sol is the natural first candidate to test.

Is GPT-5.5 better than Claude Opus 4.7?

In the table OpenAI published at launch, GPT-5.5 wins on coding in the terminal (82.7% against 69.4%), office work (84.9% against 80.3%) and the MRCR v2 long-context test (74.0% against 32.2%), and loses on SWE-Bench Pro (58.6% against 64.3%), MCP Atlas and Humanity's Last Exam. The independent scores are at the top of this page.

What is the difference between GPT-5.5 and GPT-5.5 Pro?

According to OpenAI's safety report, Pro is the same underlying model using parallel compute at answer time. It costs US$ 30 and US$ 180 per million tokens, six times GPT-5.5, has no cache discount and only works through the Responses API.

Where does GPT-5.5 rank among AI models?

In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, GPT-5.5 ranks #15 out of 267 models, with an ECI of 158.95. 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-5.5 better than Claude Haiku 5.5?

Across the 16 tests both models took and Epoch AI compiled, GPT-5.5 comes out ahead on 11, behind on 5 and tied on 0. 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-5.5 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-5.5 are: FrontierMath (#1 of 37); Terminal-Bench (#1 of 31); Cl Bench Life (#1 of 13). 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-5.5 announcement
  2. OpenAI, GPT-5.5 model page (documentation)
  3. OpenAI, API pricing
  4. OpenAI, GPT-5.5 system card
  5. OpenAI, GPT-5.5 usage guide and GPT-6 migration
  6. OpenAI, Codex models and GPT-5.5 retirement
  7. OpenAI, API deprecations
  8. OpenAI, API changelog
  9. OpenAI, prompt caching (model differences)
  10. OpenAI, data residency (regions and models)
  11. OpenAI, GPT-5.6 announcement

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