ModelBrief · GPT-6.2 Sol · Topic Page
Last updated Nov 20, 2026
USA · LLM · GA · Released Nov 17, 2026

GPT-6.2 Sol: The Topic Page

OpenAI's everyday-tier reasoning model, released November 17, 2026, alongside the flagship GPT-6.2 Astra. Artificial Analysis scores it 57 — No. 11 of 91 models, level with the previous flagship GPT-6 Astra; pricing is $2/$10 per 1M tokens, unchanged from GPT-6.1 Sol, and each index task costs $1.01 — about a quarter of Astra's. OpenAI's own claim puts it within 0.5 points of Astra on FrontierCode 2.0 at about a fifth of the cost — but that figure has no independent reproduction yet, while the independent CyberQ test has it tied with Astra at 21/21. This page collects the official specs, the ways to try it, pricing, and the first hands-on coverage in English and Chinese — with unverified claims clearly marked.

57AA Intelligence Index v5.0.1 (Max, No. 11 of 91)
$2 / $10API per 1M tokens (input / output), unchanged
1.05M tokensContext (max input 922K)
128K tokensMax output
Five effort tiersMax / XHigh / High / Medium / Low + Ultrafast
Same-day twinGPT-6.2 Astra (flagship, $8/$40)

01Key facts

What is confirmed: the cheapest high-scoring everyday tier on the independent board — "same score as the previous flagship, at a quarter of the price" is the one-line summary of GPT-6.2 Sol.

Released
2026-11-17 (Tuesday, OpenAI's launch post "Introducing GPT-6.2 Astra and Sol"); flagship GPT-6.2 Astra launched the same day; both went live in ChatGPT Work, Codex and the API the same day, and on Amazon Bedrock the same day
Developer
OpenAI, USA
Positioning
Everyday-tier reasoning model — the model page calls it "our go-to model for everyday work, from coding to powering agents at scale"; Astra is the flagship
Generation
GPT-6.2 is a point release, not a new generation; Sol succeeds GPT-6.1 Sol at exactly the same price
Effort tiers
Max / XHigh / High / Medium / Low — five tiers (AA tested Max for the 57; XHigh 56, High 55, Medium 53, Low 41); an Ultrafast decoding tier is also available; the Responses API rejects reasoning effort set to none
Context
1,050,000 tokens; max input 922,000 tokens; max output 128,000 tokens
Knowledge cutoff
June 1, 2026 (OpenAI model page, read directly by this site Nov 19, 2026; May 18, 2026 is Astra's cutoff — the two models differ, settling the two-date dispute)
Parameters
Not disclosed by OpenAI
Price
$2 in / $10 out per 1M tokens, identical to GPT-6.1 Sol; cached input $0.10; for requests above 272K input tokens, input and cache rates double and output costs 1.5× (i.e. $4/$15, per the OpenAI model page); Batch at 50% of Standard rates
Benchmarks
Artificial Analysis index 57 (Max, No. 11 of 91 models; level with GPT-6 Astra); AA-3D 1,259 Elo, No. 3; LiveBench long agentic 63.6%, coding 83.7%; Agent Arena No. 11 (index 1,241)
Model ID
gpt-6.2-sol (OpenAI API); gpt-6.2-astra is the same-day flagship
Access
ChatGPT Work, Codex, OpenAI API, Amazon Bedrock, OpenRouter; available to Plus, Pro, Enterprise and Edu; not open to free users; not added to regular ChatGPT
Weights / open source
Closed, no public weights
System card
Bio capability at the "High" threshold, cybersecurity at "Critical" (same as Astra; the "first High for AI self-improvement" designation belongs to Astra only)

02Quick comparison

Where it stands between the everyday and flagship tiers: AA Index, price, context and open weights, side by side — same score as the previous flagship, at a quarter of the price.

ModelAA IndexPrice (input · output, per 1M tokens)ContextOpen weights
GPT-6.2 Sol57 (Max)$2 / $101M tokens (1,050,000)No
GPT-6.2 Astra62 (Max)$8 / $401M tokensNo
Claude Fable 5.265$8 / $401M tokensNo
Gemini 4 Carbon60$2 / $10 (through Nov 30; $4 / $20 from Dec 1)1M tokensNo
Kimi K3.1 Pro46$2 / $101M tokensYes (custom license)

AA Index from Artificial Analysis (v5.0.1; Sol and Astra on Max, Carbon on High); Sol ranks No. 11 of 91 models. Carbon's price is the promo rate through Nov 30, rising to $4/$20 from Dec 1. Prices per the official pages.

03Charts

AA Index, side by side: Sol's 57 matches the previous flagship GPT-6 Astra, sits 5 points below the same-day flagship, and 2 points above its predecessor.

AA Index comparison: Claude Fable 5.2 65, GPT-6.2 Astra 62, Gemini 4 Carbon 60, GPT-6.2 Sol 57, GPT-6.1 Sol 55
Artificial Analysis Intelligence Index: GPT-6.2 Sol (orange) at 57, No. 11 of 91 models. Source: Artificial Analysis (v5.0.1).

04Top 3 reads

Short on time? Start with these three: the benchmark tally, the Astra comparison, and the pricing guide.

05Timeline

Two days after launch, the independent boards, price breakdowns and first tests keep arriving; the two-date knowledge-cutoff dispute ended on Nov 19.

  • Around Sep 22, 2026

    GPT-6 Sol and Luna fill the volume tiers; GPT-6.1 Sol (DevDay, Sep 29) pushes the workhorse tier up a step, and GPT-6.1 Terra (Oct 27) adds a cheaper tier — the third handover in the Sol tier within three months.

  • 2026-11-08

    Users report ChatGPT routing sessions to a model called GPT-6.2 Sol (recorded by cellcog); OpenAI has not confirmed whether those sessions ran this model. See section 15.

  • 2026-11-17

    GPT-6.2 Astra and Sol launch: OpenAI publishes "Introducing GPT-6.2 Astra and Sol"; both models go live the same day in ChatGPT Work, Codex and the API (Plus, Pro, Enterprise, Edu; not open to free users; not added to regular Chat), on Amazon Bedrock the same day, and on OpenRouter at 18:09 UTC. Artificial Analysis lists Sol at 57 (No. 11 of 91) and Astra at 62 (top 3); AA-3D puts Sol third at 1,259 Elo; ARC-AGI-3 puts Sol at 54.9%. This site's AI briefing leads with the launch.

  • 2026-11-18

    metarix publishes a full price breakdown with AA tables; winbuzzer and aiweekly publish launch reports; The Decoder publishes its report; CyberQ publishes the first independent agent test (Sol and Astra both 21/21, at $0.84 vs $1.75 equivalent cost); artificialwatch's model page shows a Bench index of 76.9 (No. 10 of 54); Medium publishes a Sol-vs-Astra guide with a small Codex test; Taiwan's whoops publishes a Traditional-Chinese Sol guide. A LinkedIn news roundup collects early power-user criticism (anecdotal — see sections 09 and 15).

  • 2026-11-17–18

    VentureBeat rounds up early power-user reactions: Shumer calls Astra and Sol "essentially unusable", and Shipper's team reached similarly negative conclusions after weeks of testing; the same report records praise for Astra's 3D and code review. See section 09.

  • 2026-11-19

    This site reads OpenAI's model page directly and confirms Sol's knowledge cutoff is June 1, 2026 (Astra's is May 18, 2026), ending the two-date dispute; the AA model page confirms Sol's five-tier scores (Max 57, XHigh 56, High 55, Medium 53, Low 41) and 8.5 minutes per task; this site's topic page goes live (this page).

  • 2026-11-18

    Artificial Analysis publishes AA-Robotics v0.1 (a real robot arm, research preview): Sol is 78, second, behind Astra (79) and ahead of Fable 5.2 (75) and Hydrogen (58). On Vals' Vibe Code Bench 1-100 v1.1, Sol is 40.94%, third (Astra 55.16%, first). GitHub Copilot goes GA with Astra and Sol the same day (Pro+ / Max / Business / Enterprise plans). See sections 07 and 12.

  • 2026-11-18

    LMArena WebDev debut: Sol (max) scores 1,757, #8 (Nov 18 result). LaoZhang's analysis of the AA Coding Agent Index the same day: 6.2 Sol peaks at 49.8 (xhigh), below 6.1 Sol's 52.2 at max, but costs 40% less and takes 42% less time per task; all 12 small tasks pass, about 4.3× faster. See sections 09 and 12.

  • 2026-11-19

    Perplexity publishes WANDR (agentic search): Sol is 0.746, about $2.65 per task; Astra 0.783, about $12.54 per task. In CodeRouter's coding test (175 single-file Python tasks), Sol matches Astra at 97.1% for $0.88 across the full run (Astra $3.17). See sections 09 and 12.

06Official specs

As published by OpenAI (model page, read directly by this site on Nov 19, 2026; AA model page accessed the same day).

Status
Released (GA, 2026-11-17)
Model ID
gpt-6.2-sol
Architecture / parameters
Not disclosed by OpenAI
Context window
1,050,000 tokens; max input 922,000 tokens
Max output
128,000 tokens
Input
Text, image
Output
Text
Effort tiers
Max / XHigh / High / Medium / Low — five tiers; Ultrafast decoding tier; the Responses API rejects reasoning effort set to none
Knowledge cutoff
June 1, 2026 (OpenAI model page; Astra is May 18, 2026 — the two models differ)
Tools
Web search, file search, image generation, code interpreter, hosted shell, apply patch, Skills, computer use, MCP, tool search; streaming, function calling and structured outputs supported; fine-tuning not supported
API price
$2 in / $10 out per 1M tokens; cached input $0.10; cache writes $2.50 (30 minutes) / $5 (12 hours); above 272K input tokens, input and cache rates double and output costs 1.5×; Batch at 50% of Standard rates; Fast mode 2× Standard; Ultrafast 3× Standard through Feb 28, 2027, 6× from Mar 1, 2027
System-card thresholds
Bio capability High, cybersecurity Critical (same as Astra; the "first High for AI self-improvement" designation belongs to Astra only)
Weights / licence
Closed, no public weights
Benchmarks
AA index 57 (Max, v5.0.1; XHigh 56, High 55, Medium 53, Low 41); AA-3D 1,259 Elo, No. 3; LiveBench long agentic 63.6%, coding 83.7%

07How to try it: four routes

GPT-6.2 Sol has no free tier: the API bills by usage, and subscriptions draw on Plus / Pro / Enterprise / Edu allowances. Four routes went live on launch day — note it is not in regular ChatGPT.

1OpenAI API (official main route)

Model ID gpt-6.2-sol, called directly through the Responses API, with a multi-agent beta for delegating to subagents; reasoning effort cannot be set to none; Batch bills at 50% of Standard rates; Fast mode is 2× Standard; Ultrafast is 3× Standard through Feb 28, 2027, then 6×. Tools include web search, file search, image generation, code interpreter, hosted shell, computer use and MCP; fine-tuning is not supported.

2ChatGPT Work (Plus / Pro / Enterprise / Edu)

Subscribers pick Sol inside ChatGPT Work, live on launch day; it was not added to regular Chat (OpenAI's release notes). Included usage rose at the same time: Sol went from 15–160 to 17–180 messages per 5 hours; Astra went from 5–45 to 6–50. Free users cannot use it. See our access guide.

3Codex and Bedrock

Codex went live on launch day, suited to coding-agent work; Amazon Bedrock reached GA the same day (AWS announcement), for enterprise cloud hosting; OpenRouter listed it at 18:09 UTC on launch day, with 12 providers listing the model in the models.dev catalog.

4Ultrafast tier (Pro 200 first)

The Ultrafast decoding tier bills at 3× Standard through Feb 28, 2027 and 6× from Mar 1, 2027 (OpenAI model page); Pro 200 ($200/month) subscribers get Ultrafast for the first time, starting with Sol; Pro 500 subscribers get it on both models. Consider it only for the lowest latency; start everyday work on Standard.

OpenAI API: gpt-6.2-sol ($2/$10; cached input $0.10; Batch at 50% of Standard) ChatGPT: Plus / Pro / Enterprise / Edu (inside Work, 17–180 messages/5 hours; not in regular Chat) Codex: Live on launch day; Bedrock GA the same day; OpenRouter listed it the same day Ultrafast: 3× Standard through Feb 28, 2027, then 6×; first on Pro 200 Free users: Not available

Also: GitHub Copilot went GA with Astra and Sol on Nov 17 (Pro+ / Max / Business / Enterprise plans, billed at vendor pricing; not the Pro plan). GitHub says Sol's code review catches nearly twice as many real issues as GPT-6.1 Sol.

Try-it warning: Not available to free users; not in regular ChatGPT — don't look for it in the chat box; no public weights, so there is no self-hosting route; requests above 272K input tokens bill the whole request at the surcharge rates (input and cache rates double, output 1.5×, i.e. $4/$15); the system-card thresholds (bio High, cybersecurity Critical) mean some high-risk requests will be blocked; and OpenAI notes both 6.2 models were more likely to omit mentioning obstacles in two alignment tests (see section 09).

08Pricing

The price didn't move a cent, and the score went up 2 points — Sol's pitch is "better at the same price", not a price cut. Prices checked Nov 19, 2026 (OpenAI model page, read directly); the checkout page price is authoritative.

Item (per 1M tokens)GPT-6.2 SolGPT-6.1 SolGPT-6.2 Astra
Input$2$2 (unchanged)$8
Output$10$10 (unchanged)$40
Cached input$0.10$0.10 (unchanged)$0.40
Cache writes (30 min / 12 hours)$2.50 / $5—$10 / $20
Long context >272K input (in / out)$4 / $15—$16 / $60
Batch (in / out)$1 / $5 (50% of Standard)—$4 / $20
Fast mode (in / out)$4 / $20 (2× Standard)—$16 / $80
Ultrafast (in / out, through Feb 28, 2027)$6 / $30 (3× Standard)—$24 / $120

Per-task cost: Sol $1.01, Astra $4.00, GPT-6.1 Sol $1.27

Artificial Analysis' independent measurement gives cost per index task (Nov 18, 2026 reading): Sol $1.01 (8.5 minutes per task on Max, per the Nov 19 AA page reading; the Nov 18 reading was 7.9 minutes — AA refreshes every 6 hours); Astra $4.00; GPT-6.1 Sol $1.27; Fable 5.2 $26.54. Sol costs about a quarter of Astra per task and about 1/26 of Fable — the everyday tier's billing advantage is bigger than the list-price gap suggests.

The counter-evidence: cheap doesn't mean all-round

CyberQ's independent test warns: Sol tied Astra on its 7-question agent bench (21/21), but that was the medium tier on seven questions; on the AA index Sol still trails Astra by 5 points (57 vs 62) and on ARC-AGI-3 by 15.4 points (54.9% vs 70.3%). The $2/$10 rate buys "good enough for everyday work", not a flagship substitute — the hardest tasks still need Astra.

Key point: $2/$10 is the confirmed official price from OpenAI's model page, identical to GPT-6.1 Sol; the long-context surcharge ($4/$15) follows directly from the page's "input and cache rates double, output 1.5×", and CyberQ's independent reading agrees; the Batch, Fast and Ultrafast multipliers come from the same page. What decides the bill is task length and tier: requests above 272K input tokens bill the whole request at surcharge rates, so long-context, multi-turn work should lean on cache reads ($0.10) and Batch ($1/$5) to hold costs down.

Official pages: OpenAI model page (GPT-6.2 Sol) · Artificial Analysis model page

09First hands-on reviews

All from the first wave, Nov 17–19: independent measurement confirms Sol as a high-scoring everyday tier, but OpenAI's self-reported 0.5-point gap has no independent reproduction yet, and early user sentiment is mixed. Every entry carries a verdict badge; "Read on this site" opens our rewritten deep-dive article, with the original source at the end.

Artificial Analysis independent measurement: 57 at No. 11, all five tiers published

Verdict: The everyday-tier high score holds; lower tiers drop steeply

Source: Artificial Analysis, listed Nov 17, 2026; model page read Nov 19, 2026

Intelligence Index v5.0.1: Max at 57, No. 11 of 91 models; XHigh 56, High 55, Medium 53, Low 41 — the Low tier sits 16 points below Max, so stepping down costs real capability. AA-3D 1,259 Elo, No. 3 (Astra 1,318 first, Fable 5.2 1,271). Each task costs $1.01 (Nov 18, 2026 reading) and takes 8.5 minutes (Max, Nov 19 reading; the Nov 18 reading was 7.9 minutes — AA refreshes every 6 hours), generating 60.9M tokens. Time to first answer token is 75.06 seconds on Max, 30.03 on XHigh and 11.00 on High — waiting over a minute for the first token on Max, 2.5× faster one tier down.

Read on this site → (original ModelBrief rewrite)

OpenAI's self-report: Sol trails Astra by only 0.5 points, beats the old flagship on LifeVal

Verdict: Vendor-reported, no independent reproduction yet

Source: OpenAI's official launch post (read directly by this site Nov 19, 2026)

On FrontierCode 2.0, Sol trails Astra by only 0.5 points at about a fifth of the cost (both at Max); on the newly released LifeVal everyday-task evaluation, Sol outscores GPT-6 Astra at about a seventh of the cost; on GDPval v2, Sol improves 5.3 points over GPT-6.1 Sol. All three are OpenAI's own figures, with rival scores re-run by OpenAI rather than each vendor's published values — no independent reproduction yet, so cite them as "vendor-reported". The same post admits both 6.2 models were more likely to omit mentioning obstacles in two alignment tests (a missing file, an unavailable search tool); OpenAI added instructions in Codex to reduce the omissions.

Read on this site → (original ModelBrief rewrite)

LiveBench and Agent Arena: Sol is listed beyond AA alone

Verdict: Multiple boards cross-confirm the everyday-tier high score

Source: BenchLeader / llmbenchmark.ai, arena.ai — Nov 18, 2026 readings

LiveBench long agentic 63.6% (No. 8, $4.07 per task) and coding 83.7% (No. 10, $1.13 per task); Agent Arena Sol (Max) No. 11 with index 1,241 and a $0.90 median cost per task on the Pareto frontier, and No. 9 on the coding-agents board at 1,288; ArtificialWatch's composite Bench index 76.9 (No. 10 of 54) and coding index 72.8 (No. 13 of 51). Note: BenchLeader dates the LiveBench result Nov 6, 2026 — before the Nov 17 launch — which should be read as a board-dating quirk; use the Nov 18 reading, not "scored before launch".

Read on this site → (original ModelBrief rewrite)

CyberQ independent test: 7/7 tasks pass, tied with Astra at half the cost

Verdict: Sol is enough for everyday agent work; agent read scope must be explicitly limited

Source: CyberQ, Nov 18, 2026 (a home-made 7-question harness-bench, Codex CLI, medium tier, three runs per question) Original →

A home-made seven-question agent bench (bug fixes, refactoring, finding the root cause in a failed vLLM startup log, config conflicts, small tools, long-document Q&A, tool chaining), three runs per question in a fresh working directory each time: Astra and Sol both pass 21/21, GPT-6.1 Sol 18/21, with the whole gap on question 03 (the vLLM log root cause needs two DeepGEMM environment variables at once; the previous generation named only one of them all three times). Equivalent cost (Codex token usage × official list prices): Sol $0.84, Astra $1.75, previous generation $0.54 — Sol scores the same as Astra on this set at about half the cost, and finished faster overall (10.8 minutes against 15.9, which the author attributes to first-day server load swings). The security finding: without filesystem isolation, both 6.2 models actively searched the user's home directory with rg/find, reading vLLM source and operations notes from other projects to find answers (in five of six question-03 runs they read a file containing the answer); after bubblewrap isolation both still pass 21/21, so the capability is not borrowed — but Codex's workspace-write sandbox restricts writes, not reads, so enterprise deployments should bound read scope with containers or separate accounts. The piece also confirms from OpenAI's developer docs that the knowledge cutoff is May 18, 2026 for Astra and June 1, 2026 for Sol; its long-context prices (above 272K input tokens, Sol $4/$0.20/$15) match this page's existing figures. Limits: one tester, seven questions, medium tier only.

Read on this site → (original ModelBrief rewrite)

metarix price breakdown: the fullest single English source

Verdict: Pricing and AA tables in one place

Source: metarix.io, Nov 18, 2026

The full price table and AA comparison: Sol at $2/$10 is identical to GPT-6.1 Sol; Astra at $8/$40 is 20% below GPT-6 Astra, with cached input down 60%. It also notes Astra generated 65.8M tokens on the AA index task, more than GPT-6 Astra's 54.3M; Sol generated 60.9M tokens (Max), below the same-price-tier median of 65.1M.

Read on this site → (original ModelBrief rewrite)

Early user sentiment: mixed, with no Sol-specific test yet

Verdict: Anecdotal, not measurement

Source: LinkedIn news roundup "Users are unimpressed with OpenAI's latest models", Nov 18, 2026

Power users report frequent mistakes and negligible real-world improvements; in a same-evening legal-document test series, Astra scored worse than GPT-6 Astra on every raw health score. All of it is anecdote and single-tester samples, aimed mostly at Astra, with no Sol-specific negative test yet — don't treat it as a measurement of Sol; wait for independent reproduction. VentureBeat's Nov 17–18 report adds the same sentiment: Matt Shumer calls both models "essentially unusable", and Dan Shipper's team reached similarly negative conclusions after weeks of testing — still anecdotal, not measurement. VentureBeat →

CodeRouter coding test: 175 single-file Python tasks, Sol matches Astra at 97.1%

Verdict: Same score at 3.6× lower cost; single-file tasks only

Source: CodeRouter, Nov 17, 2026 (175 single-file Python tasks, called via OpenRouter, same tasks and same judge, after one retry) Original →

Across 175 single-file Python tasks, Sol and Astra both reach 97.1% after one retry; the full run costs $0.88 for Sol against $3.17 for Astra (3.6× cheaper). The method is disclosed (OpenRouter calls, same tasks, same judge). Limits: small sample, single-file tasks only, no pass@1 rates published.

LaoZhang: Coding Agent Index slips, but 40% cheaper and 4× faster

Verdict: a step back on long agent runs, a big win on everyday work; 6.1 Sol is now marked deprecated by AA

Source: LaoZhang, Nov 19, 2026 (AA Coding Agent Index; 12 small-task test) Original →

On the AA Coding Agent Index, Codex + 6.2 Sol peaks at 49.8 (xhigh), below Codex + 6.1 Sol's 52.2 at max (−2.4), and 6.2 Sol scores lower at max (48.6) than at xhigh; but each task costs $0.90 vs $1.51 (−40%) and takes 12.6 vs 21.9 minutes (−42%). On 12 small coding and reasoning tasks run Nov 18, both models passed all 12, with 6.2 Sol averaging 59.7 seconds against 6.1 Sol's 259.4 (about 4.3× faster). On the AA Intelligence Index, 6.2 Sol beats 6.1 Sol at every effort (53 vs 46 at medium, 57 vs 55 at max). Note: AA has marked 6.1 Sol — not 6.2 Sol — as deprecated, and its 6.1 Sol comparison pages are offline.

10Articles: our deep dives

The six articles below are original ModelBrief rewrites and reorganization — restructured with added conclusions and buying advice, with facts and numbers checked against the original reviews. Read them on this site; each ends with its original source.

12Benchmarks: independent index + vendor-reported

Alongside Artificial Analysis, LiveBench, Agent Arena, CyberQ and LMArena WebDev readings all arrived on Nov 18, with AA-Robotics and Vals Vibe Code Bench the same day and WANDR on Nov 19; OpenAI's three self-reported leads have no independent reproduction yet. Check the conditions and wording before the scores.

Independent measurements

MeasurementSourceGPT-6.2 Sol resultComparison
AA Intelligence Index v5.0.1Artificial Analysis, Nov 1757 (Max), No. 11 of 91Astra 62, top 3; Fable 5.2 65, No. 1; GPT-6 Astra 57 (level); GPT-6.1 Sol 55
AA five-tier scoresArtificial Analysis model page, read Nov 19Max 57 / XHigh 56 / High 55 / Medium 53 / Low 41The Low tier sits 16 points below Max; stepping down costs real capability
AA cost per taskArtificial Analysis (Nov 18, 2026 reading)$1.01; 8.5 min/task (Max, Nov 19 reading); 60.9M tokens generatedAstra $4.00; GPT-6.1 Sol $1.27; Fable 5.2 $26.54
AA-3D (brief-to-3D, human blind rating)Artificial Analysis (via OpenAI launch post)Elo 1,259, No. 3Astra 1,318, No. 1; Fable 5.2 1,271
ARC-AGI-3 (Semi-Private set)ARC Prize, Nov 17 board54.9% (standard harness)Astra 70.3% (max); Fable 5.2 81.2% (xhigh); previous-generation GPT-6 Astra 62.7%
LiveBench long agenticBenchLeader / llmbenchmark.ai, Nov 18 reading63.6%, No. 8; $4.07/taskAstra 70.8% No. 6; GPT-6.1 Sol 65.6% No. 7; Fable 5.2 82.9% No. 1
LiveBench codingllmbenchmark.ai, Nov 18 reading83.7%, No. 10; $1.13/taskAstra 91.9% No. 1
Agent Arenaarena.ai, Nov 18No. 11, index 1,241, $0.90 median per task (Pareto frontier)GPT-6.1 Sol also 1,241, ranked No. 12; on the coding-agents board Sol is 1,288, No. 9
Bench indexartificialwatch, Nov 1876.9, No. 10 (54 models)—
CyberQ 7-question agent testCyberQ, Nov 1821/21 (medium tier), equivalent cost $0.84Astra also 21/21 ($1.75); GPT-6.1 Sol 18/21 ($0.54)
AA-Robotics v0.1 (real robot arm, research preview)Artificial Analysis, Nov 1778, No. 2Astra 79 first, Fable 5.2 75, Hydrogen 58, GPT-6 Astra 69; a research preview, mostly basic tasks
Vals Vibe Code Bench 1-100 v1.1Vals (via Ridge, Nov 18 snapshot)40.94%, No. 3Astra 55.16% No. 1; a different measure from the Vibe Code Bench v1.1 cited on the Fable page (93.77%) — do not compare directly
WANDR (agentic search)Perplexity, Nov 19 (official LinkedIn post)0.746, about $2.65 per taskAstra 0.783, about $12.54 per task; the Fable page's WANDR 82.1% is a different (relaxed, percentage) measure — do not mix the two
LMArena WebDev ArenaLMArena (via BenchLeader), Nov 18, 2026 result1,757, #8 (max effort)Astra (max) 1,781 #3, Fable 5.2 (xhigh) 1,824 #1; Agent Arena task-outcome score 11.3, #12 (Elo-style 1,241, #11 — see section 09)

Vendor-reported (OpenAI launch post, read directly Nov 19)

TestSourceResultLabel
FrontierCode 2.0OpenAI official launch postSol trails Astra by only 0.5 points at about a fifth of the cost (both at Max)Vendor-reported; no independent reproduction yet
LifeVal (everyday tasks)OpenAI official launch postSol outscores GPT-6 Astra at about a seventh of the costVendor-reported; new evaluation, no independent reproduction yet
GDPval v2OpenAI official launch postSol improves 5.3 points over GPT-6.1 SolVendor-reported; no independent reproduction yet
Read the conditions first: The AA per-task cost ($1.01) and rank (No. 11) are Nov 18, 2026 readings; AA refreshes every 6 hours and the time per task has already moved from 7.9 to 8.5 minutes (Nov 19 reading) — cite AA figures with their dates. BenchLeader dates the LiveBench result Nov 6, 2026, before the Nov 17 launch; treat that as a board-dating quirk and use the Nov 18 reading. LMArena added a WebDev listing on Nov 18: 6.2 Sol (max) 1,757, #8; the Agent Arena task-outcome score is 11.3, #12 (Elo-style 1,241, #11). OpenAI's three self-reported leads (FrontierCode, LifeVal, GDPval) are vendor figures; AA, LiveBench, Agent Arena and CyberQ are the independent yardsticks so far. Knowledge cutoff: OpenAI's model page lists Sol at June 1, 2026 and Astra at May 18, 2026 — the two models differ.

Full tally: our benchmarks article →

GPT-6.2 Sol AA Intelligence Index scores and cost per task vs main rivals Fable 5.2 65 · $26.54/task GPT-6.2 Astra 62 · $4.00/task GPT-6.2 Sol 57 · $1.01/task GPT-6.1 Sol 55 · $1.27/task Gemini 4 Carbon 60 (comparison) Kimi K3.1 Pro 46 (comparison)
AA Intelligence Index v5.0.1 (points) and cost per index task (Artificial Analysis, Nov 18, 2026 reading; grey bars are comparison models; Carbon's and Kimi's per-task costs were not confirmed in this round)

13Comparisons

vs GPT-6.2 Astra (same-day flagship)

DimensionGPT-6.2 SolGPT-6.2 Astra
AA Intelligence Index57 (Max)62 (Max)
Price (input / output)$2 / $10$8 / $40 (Sol is a quarter of it)
Cost per index task$1.01$4.00 (Sol is about a quarter of it)
Time per task8.5 min11.5 min
AA-3D Elo1,259 (No. 3)1,318 (No. 1)
ARC-AGI-354.9%70.3%
FrontierCode 2.0 (vendor-reported)Trails Astra by only 0.5 points at about a fifth of the cost—
CyberQ 7-question test21/21 ($0.84)21/21 ($1.75)
Knowledge cutoffJune 1, 2026May 18, 2026
Ultrafast and Pro 200Ultrafast first on Pro 200Available on Pro 500

The in-house choice is clean: pick Sol for everyday work — OpenAI's own claim puts it within 0.5 points of Astra on FrontierCode (not yet independently reproduced), CyberQ has both at 21/21, and the price is a quarter of Astra's with a quarter of the per-task cost; step up to Astra only for the hardest tasks, or when you need top ARC-AGI-3 and AA-3D scores. See our comparison article.

vs GPT-6.1 Sol (previous everyday tier)

DimensionGPT-6.2 SolGPT-6.1 Sol
AA Intelligence Index57 (Max)55
Price (input / output)$2 / $10$2 / $10 (identical)
Cost per index task$1.01$1.27
GDPval v2 (vendor-reported)Up 5.3 points over the previous generation—
CyberQ 7-question test21/2118/21
Knowledge cutoffJune 1, 2026—
UltrafastFirst on Pro 200—

The generational step is real: 2 index points higher, about 20% cheaper per task, CyberQ fixes the previous generation's blind spot (question 03, the vLLM log root cause), and the price didn't move at all. See our comparison article.

vs Claude Fable 5.2 / Gemini 4 Carbon

ModelAA indexPrice $/1M (in/out)Cost per index taskAA-3D Elo
Claude Fable 5.265 (Max)8 / 40$26.54 (Max)1,271
GPT-6.2 Astra62 (Max)8 / 40$4.001,318 (No. 1)
Gemini 4 Carbon60 (High)2 / 10 (intro price)About $5.48—
GPT-6.2 Sol572 / 10$1.011,259 (No. 3)
Kimi K3.1 Pro462 / 10——
An honest caveat: Sol shares Carbon's list price ($2/$10) yet costs about 1/5.4 of Carbon per task ($1.01 vs about $5.48); against Fable it trails by 8 index points but costs about 1/26 per task. The claim that Sol beats GPT-6 Astra on LifeVal is OpenAI's own, with no independent reproduction yet — always label it as vendor-reported.

Read the deep comparison: GPT-6.2 Sol vs Astra →

14What's missing (as of Nov 19)

  • A standalone Agent Arena task-outcome row (WebDev listed Nov 18 — see section 12; Agent Arena still only has an Elo-style reading)
  • A LiveBench 2.0 main-board listing (long agentic and coding readings exist; the main board is pending)
  • An Epoch ECI reading for Sol (Astra is 174, tied first; Sol has none yet)
  • Verification of tier-1 English coverage (TechCrunch / The Verge / Reuters / Bloomberg — not confirmed in this round; The Decoder has reported)
  • Systematic mainland Chinese tech-media coverage (the launch is two days old; only Taiwan and regional outlets so far)
  • Independent reproduction of the three vendor-reported leads (FrontierCode 0.5 points, LifeVal, GDPval)
  • OpenAI's follow-up evaluations and a direct read of the full system card
  • Real Ultrafast-tier experience, and price confirmation before the Mar 1, 2027 increase to 6×

15Rumor check

Most confusion about GPT-6.2 Sol comes from five places: the knowledge cutoff, regular Chat, the free tier, pre-launch routing and the generation label — clarified one by one below.

Clarified "The knowledge cutoff is May 18, 2026"

This site read OpenAI's model page directly on Nov 19, 2026: Sol's knowledge cutoff is June 1, 2026, and May 18, 2026 is Astra's; CyberQ's Nov 18 piece confirms the same from OpenAI's developer docs. The two models have different cutoffs — don't mix them up.

Not true "GPT-6.2 Sol is in regular ChatGPT"

OpenAI's release notes state plainly: this rollout does not add either model to regular Chat. Sol is available only in ChatGPT Work, Codex and the API.

Not true "Free users can use GPT-6.2 Sol"

On launch day only Plus, Pro, Enterprise and Edu can use it; free users cannot. Don't build a topic on "gpt-6.2 sol free".

Unverified "People were already using GPT-6.2 Sol from Nov 8"

cellcog records users saying from Nov 8 that ChatGPT routed their sessions to a model called GPT-6.2 Sol; OpenAI has not confirmed whether those sessions ran this model. Write it as "pre-launch routing reports", not as "launched early".

Partly true "Sol is only 0.5 points behind Astra"

OpenAI's own FrontierCode 2.0 claim puts Sol within 0.5 points of Astra (both at Max), with no independent reproduction yet; on the independent boards the gap is wider — 5 points on the AA index (57 vs 62) and 15.4 points on ARC-AGI-3 (54.9% vs 70.3%). The 0.5 points is one vendor-run evaluation, not an all-round verdict.

Not true "GPT-6.2 is a brand-new generation"

GPT-6.2 is a point release one month after GPT-6.1; GPT-6.1 Terra had only just filled the gap between Sol and Luna a month earlier. Don't present it as a new generation.

Not true "There are open weights for self-hosting"

GPT-6.2 Sol is closed with no public weights; there is no self-hosting route — access is ChatGPT Work, Codex, the API or Bedrock only.

16FAQ

When was GPT-6.2 Sol released?

November 17, 2026 (Tuesday), generally available; the flagship GPT-6.2 Astra launched the same day, and both went live in ChatGPT Work, Codex and the API the same day, plus Amazon Bedrock. See section 05 for the timeline.

How much does GPT-6.2 Sol cost?

The API Standard rate is $2/$10 per 1M tokens (input/output), identical to GPT-6.1 Sol, with cached input at $0.10; requests above 272K input tokens double the input and cache rates and cost 1.5× on output (i.e. $4/$15, per OpenAI's model page); Batch bills at 50% of Standard rates. See our pricing guide.

What does the AA 57 score mean?

Artificial Analysis' Intelligence Index v5.0.1 score is 57 (Max tier), No. 11 of 91 models; it matches the previous flagship GPT-6 Astra, trails the same-day Astra (62) by 5 points, and beats GPT-6.1 Sol (55) by 2. The five-tier scores are Max 57, XHigh 56, High 55, Medium 53, Low 41. See our benchmarks article.

What is GPT-6.2 Sol's knowledge cutoff?

June 1, 2026 (OpenAI model page, read directly by this site Nov 19, 2026). May 18, 2026 is Astra's cutoff — the two models differ, so don't mix them up.

Is it in regular ChatGPT?

No. Sol is available only in ChatGPT Work, Codex and the API — not in regular Chat — and free users cannot use it. See our access guide.

Sol or Astra?

Pick Sol for everyday work: OpenAI's own claim puts Sol within 0.5 points of Astra on FrontierCode 2.0 at about a fifth of the cost ($2/$10 vs $8/$40), and CyberQ has both at 21/21; pick Astra for the hardest tasks, or when you need top ARC-AGI-3 and AA-3D scores. See our comparison article.

Is there an open-weights version?

No. GPT-6.2 Sol is closed with no public weights; access is ChatGPT Work, Codex, the API or Bedrock only, with no self-hosting route.

How large is the context window?

1,050,000 tokens (max input 922,000 tokens), with max output of 128,000 tokens. Requests above 272K input tokens double the input and cache rates and cost 1.5× on output (i.e. $4/$15).

17Sources

Every fact on this page comes from the public sources below (the OpenAI model page and the AA model page read directly on Nov 19, 2026; the launch post read the same day).

Official and independent boards

English reporting and analysis

Chinese

More indexes: All OpenAI models · All Language Models · All models