Tabbit
ResourcesBlogModels
Tabbit LogoTabbit

Tabbit — The AI Browser that Works for You

Topics

  • Best Browsers 2026: Rankings and Top Lists
  • AI Browser Guides and Resources
  • Browser Comparisons: Head-to-Head Guides
  • Browser Reviews and Explainers
  • Browser Alternatives
  • Browser Directory: Guides by Browser
  • Agentic Browser Resources
  • Browser Downloads and Install Guides
  • Browser Productivity Resources
  • SillyTavern Model Guides and Presets

Popular Guides

  • Best Browsers 2026: Top 10 Ranked & Compared
  • Best Free AI Browsers 2026: Plans & Prices Compared
  • Rabbit Browser: What It Is & Why Tabbit Is Better
  • Firefox vs Chrome 2026: Privacy, RAM & Speed Compared
  • Best Lightweight Browsers 2026: 7 Tested for RAM & Speed
  • AI Browser List 2026: 10 Browsers Compared
  • Neobrowser: Norton Neo, Download & Alternatives
  • Edge vs Chrome 2026: Speed, RAM & Battery Tested
  • Best Internet Browser for Android 2026
  • Best AI Browser 2026: Top 9 Tested & Ranked

Events

  • Tabbit Skill Competition
  • KPOP SBTI Fandom Personality Test
  • Tabbit Campus Creator Program
  • fifi's Picks: AI Skills for Research Papers
  • User Survey

About

  • Tabbit Blog
  • Press & Media
English
简体中文English
Review
OfficialGPT-6.2 Sol

GPT-6.2 Sol Launch Benchmarks and Cost Basis

Original source

OpenAI official blog

AuthorOpenAI

Source date2026-11-17

Tabbit curation2026-11-17

Read original

One-sentence takeaway

OpenAI says GPT-6.2 Sol scored 5.3 percentage points higher than GPT-6.1 Sol on GDPval v2. On Cognition's FrontierCode 2.0, its score was less than 0.5 points behind GPT-6.2 Astra at about one-fifth the cost (both run at Max). And on LifeVal, it scored higher than GPT-6 Astra at about one-seventh the cost. The source does not provide Sol's absolute scores, sample sizes, or full run configurations for these benchmarks, so the results cannot be independently recalculated or generalized to other tasks.

Use cases

  • Tasks these results can inform: Understanding the benchmark comparisons, relative-cost claims, and API pricing disclosed by OpenAI at launch.

  • Tasks these results should not be generalized to: Estimating actual costs for arbitrary tasks, comparing absolute capabilities across different evaluation frameworks, or inferring undisclosed benchmark details.

  • Applicable model version: GPT-6.2 Sol (API model ID: gpt-6.2-sol).

  • Test environment or client: The source only says that both sides of the FrontierCode 2.0 comparison used Max. Other test environments and clients are not specified.

  • Reasoning tier and parameters: Max for FrontierCode 2.0. Other tiers, sampling parameters, and run settings are not specified.

Evaluation method

This is a vendor benchmark summary in an OpenAI launch announcement, not an independent replication report. The source provides the following visible descriptions of the relevant evaluations:

  • FrontierCode 2.0: A Cognition benchmark that evaluates whether changes to real codebases meet maintainers' merge standards. OpenAI only says that Sol's score was less than 0.5 points behind Astra and that Sol cost about one-fifth as much. Both were set to Max.

  • LifeVal: OpenAI's everyday-task benchmark, which evaluates performance on tasks based on personal needs. The page says Sol scored higher than GPT-6 Astra at about one-seventh the cost, but does not disclose absolute scores or specific run conditions.

  • GDPval v2: An OpenAI internal benchmark covering long-horizon knowledge-work tasks in fields including finance, data science, biotechnology, and marketing. The page says Sol scored 5.3 percentage points higher than GPT-6.1 Sol. It does not disclose Sol's absolute result, sample size, or detailed protocol.

  • AA-3D: The page cites Artificial Analysis's blind-evaluation Elo results. Reviewers vote for their preferred one of two interactive works generated from the same prompt. GPT-6.2 Sol scored 1,259 Elo. The page does not list Sol's sample count, confidence interval, or full run parameters.

Other than the Max setting for FrontierCode and the blind-evaluation description for AA-3D, reasoning tiers, prompts, tools, hardware, number of runs, sample sizes, and statistical intervals are not specified.

Key results

Benchmark or basisGPT-6.2 Sol launch disclosureComparison targetConditions and limitations
FrontierCode 2.0Less than 0.5 points behind GPT-6.2 Astra. Costs about 1/5 as muchGPT-6.2 AstraBoth set to Max. Absolute scores, cost units, and sample size not disclosed
LifeValScored higher than GPT-6 Astra. Costs about 1/7 as muchGPT-6 AstraAbsolute scores, cost units, and run conditions not disclosed
GDPval v25.3 percentage points higher than GPT-6.1 SolGPT-6.1 SolAbsolute scores, sample size, and detailed test protocol not disclosed
AA-3D1,259 EloThe page also lists Astra at 1,318 and Claude Fable 5.2 at 1,271Blind preference vote using the same prompt. Sample size, intervals, and retest settings not specified
Standard API pricingInput: $2 / million tokens. Cached input: $0.10 / million tokens. Output: $10 / million tokensGPT-6.1 SolThe page says pricing carries over from GPT-6.1 Sol. Actual task costs depend on factors such as token usage

Raw data

  • OpenAI's summary for GPT-6.2 Sol says GDPval v2 improved by 5.3 percentage points over GPT-6.1 Sol. This is a vendor-reported value in the announcement. The page does not provide absolute scores or per-question data.

  • For FrontierCode 2.0, the disclosure is that Sol's score was less than 0.5 points behind GPT-6.2 Astra, and Sol cost about one-fifth as much. Both used Max. Absolute scores and the cost measurement basis are not disclosed.

  • For LifeVal, the disclosure is that Sol scored higher than GPT-6 Astra at about one-seventh the cost, without a numerical score.

  • For AA-3D, Sol is listed at 1,259 Elo. The same passage lists Astra at 1,318 and Claude Fable 5.2 at 1,271.

  • API prices per million tokens are: $2 for input, $0.10 for cached input, and $10 for output. These are unit prices, not the total task costs for the benchmarks above.

Conclusions and limitations

The narrow conclusion supported by these data is that OpenAI's launch materials report a GDPval v2 improvement for Sol over GPT-6.1 Sol and claim lower relative costs against Astra on FrontierCode and LifeVal. AA-3D gives a score of 1,259 Elo. The page does not show the raw scores, sample sizes, number of runs, or full configurations needed to recalculate most results. The cost ratios and benchmark scores should therefore be cited as OpenAI's launch claims, not treated as independent measurements or general performance guarantees. AA-3D's Elo comes from the benchmark provider's evaluation. This article records only the value cited on OpenAI's page.

Reproduction notes

The announcement omits the full test setup, per-task results, and cost breakdowns. To rerun a benchmark, use matching dataset, model, tool, and scoring versions and record task costs separately.

Curated by Tabbit

This is a third-party source navigator. Model versions, test environments, and personal experience vary; consult the original source.

GPT-6.2 Sol

Use and compare models in Tabbit

GPT-6.2 Sol

Related reviews

OfficialGoogle DeepMind2026-11-18

GPT-6.2 Sol in Google DeepMind's Gemini 4 Neon Model Card Benchmark Table

OfficialArtificial Analysis2026-11-17

GPT-6.2 Sol Scores 3 Points Lower Than Its Predecessor on the Codex Coding Agent Index

MediaAI IQ

GPT-6.2 Sol: AI IQ Subscores, Benchmark Results, and Estimation Limits

MediaArtificial Analysis

Artificial Analysis: GPT-6.2 Sol Reasoning Levels, Task Time, and Pricing

GPT-6.2 Sol

Related showcases & prompts

OfficialOpenAI API documentation

GPT-6.2 Sol Model ID and Responses API Reasoning Configuration

OfficialOpenAI API documentation

GPT-6.2 Sol API Migration and Tool Configuration

OfficialChatGPT Learn (OpenAI)2026-11-17

Task Selection and Short Workflows for GPT-6.2 Sol in Codex

OfficialOpenAI API documentation

Review PRs in Parallel with GPT-6.2 Sol in the Responses API