Official docs · as of 2026-11-18

GPT-6.2 Sol vs GPT-6.2 Astra: Official Pricing, Benchmarks & Guide (2026-11)

As of 2026-11-18, OpenAI launched both gpt-6.2-sol and gpt-6.2-astra on 2026-11-17: on OpenAI's official API pricing page for short-context requests (≤272K input), gpt-6.2-sol is $2 input / $10 output and gpt-6.2-astra is $8 input / $40 output per 1M tokens. On FrontierCode 2.0 (both at max reasoning effort), gpt-6.2-sol trails gpt-6.2-astra by less than 0.5 points, while gpt-6.2-astra takes #1 on AA-3D at 1,318 Elo and beats GPT-6 Astra by 5.2 points on Agents' Last Exam. Both models are supported on QCode using one cr_ API key across Codex and the OpenAI SDK.

Updated 2026-11-18

Switch models on the same endpoint and measure the cost per token yourself; check /models for callable models.

#$2/$10 vs $8/$40#Within 0.5 pts on FrontierCode 2.0#AA-3D 1,318 vs 1,259#1,050,000 context

Four Core Differences and Official Specs

$2 / $8

OpenAI official API Standard pricing (per 1M tokens)

Per OpenAI's official API pricing (≤272K input Standard tier, per 1M tokens): gpt-6.2-sol costs $2 input, $0.10 cached input, $2.50 for 30-minute cache writes ($5 for 12-hour), and $10 output, matching gpt-6.1-sol; gpt-6.2-astra costs $8 input, $0.40 cached input, $10 for 30-minute cache writes ($20 for 12-hour), and $40 output, 20% lower than gpt-6-astra's $10 / $50.

0.5 / 1,318

FrontierCode 2.0 and AA-3D benchmark results

OpenAI's official benchmarks show gpt-6.2-sol within 0.5 points of gpt-6.2-astra on FrontierCode 2.0 when both run at max effort; on the AA-3D leaderboard, gpt-6.2-astra ranks #1 with 1,318 Elo, ahead of claude-fable-5-2 at 1,271 Elo and gpt-6.2-sol at 1,259 Elo. On GPQA Diamond (no tools), the system card reports 97.0% for Astra and 96.0% for Sol.

2026-06-01

Knowledge cutoff and 1,050,000 context specs

Both models share identical window limits: a 1,050,000-token context window, 922,000 max input tokens, and 128,000 max output tokens, accepting text and image inputs with text output. The only spec difference is knowledge cutoff: gpt-6.2-sol is 2026-06-01, 14 days newer than gpt-6.2-astra's 2026-05-18 and more than 1 month past gpt-6.1-sol's 2026-04-30.

>272K 2x/1.5x

Long-context surcharge and reasoning_effort rule

Under OpenAI's official billing rules, when a request exceeds 272K input tokens, input and cache rates double (2x) and output costs 1.5x the short-context rate for the full request (Sol becomes $4 / $15 and Astra becomes $16 / $60). In addition, the API for both models rejects requests that set reasoning_effort="none".

Where GPT-6.2 Sol and GPT-6.2 Astra Actually Differ

As of 2026-11-18, OpenAI's official docs position gpt-6.2-sol as "Our go-to model for everyday work, from coding to powering agents at scale" and gpt-6.2-astra as "Our most capable model for your hardest projects, from start to finish." On token limits and endpoints, the two models are identical: both offer a 1,050,000 context window, 922,000 max input, and 128,000 max output; both support v1/chat/completions, v1/responses (including the Multi-agent beta for subagent delegation, computer_use, apply_patch, hosted_shell, mcp, and tool_search), and v1/batch; and both share Standard rate limits of 5,000 / 10,000 / 15,000 RPM and 1,000,000 / 4,000,000 / 40,000,000 TPM across Build, Launch, and Grow tiers. The practical trade-off is unit token cost (OpenAI official API rates of $2 / $10 vs $8 / $40) versus peak multi-step reasoning ceiling.

What Changed Moving from GPT-6.1 Sol to the 6.2 Lineup

Compared with gpt-6.1-sol (released 2026-09-29), gpt-6.2-sol (released 2026-11-17) keeps the exact same OpenAI official API unit rates ($2 input, $0.10 cached input, $2.50 30-minute cache write, $10 output per 1M tokens) while delivering four upgrades: ① knowledge cutoff advances from 2026-04-30 to 2026-06-01 (with gpt-6.2-astra at 2026-05-18); ② on FrontierCode 2.0 at max effort, it narrows the gap with flagship Astra to within 0.5 points; ③ in the Responses API, it adds Multi-agent beta subagent delegation and public-beta discounted background completion windows; ④ on ChatGPT subscriptions, Pro 200 unlocks Ultrafast mode for the first time starting with gpt-6.2-sol (while Pro 500 supports Ultrafast on both 6.2 models). Also note that OpenAI Help Center article 20001625 announced that starting December 2026, Plus, Pro, and Business usage allowances will be shared between ChatGPT and Codex.

Release Timeline

2026-09-29

OpenAI releases gpt-6.1-sol with a 1,050,000-token context window, 922,000 max input, 128,000 max output, a 2026-04-30 knowledge cutoff, and OpenAI official API rates of $2 input / $10 output per 1M tokens.

2026-11-17

OpenAI launches gpt-6.2-astra ($8 / $40 official API rate, 2026-05-18 cutoff) and gpt-6.2-sol ($2 / $10 official API rate, 2026-06-01 cutoff) across the API, ChatGPT Work, and Codex.

2026-11-18

QCode supports both gpt-6.2-astra and gpt-6.2-sol via https://api.qcode.cc/openai (Codex) and https://api.qcode.cc/openai/v1 (OpenAI SDK); all figures on this page were verified on 2026-11-18.

Confirmed vs Not Published

Confirmed (OpenAI official docs & system card)

① Model IDs & specs: Only two 6.2 API model IDs are published — gpt-6.2-astra and gpt-6.2-sol — both with a 1,050,000 context window, 922,000 max input, 128,000 max output, text and image input, and text output (no audio or video). ② OpenAI official API pricing (≤272K short-context Standard tier, per 1M tokens): gpt-6.2-astra is $8 input, $0.40 cached input, $10 30-minute cache write ($20 12-hour), $40 output; gpt-6.2-sol is $2 input, $0.10 cached input, $2.50 30-minute cache write ($5 12-hour), $10 output. ③ Speed tiers & long-context multipliers: requests above 272K input tokens pay 2x input/cache and 1.5x output for the full request; Batch is 50% of Standard, Fast is 2x, and Ultrafast is 3x through 2027-02-28 and 6x starting 2027-03-01. ④ Benchmarks & preparedness ratings: within 0.5 points on FrontierCode 2.0 (max); AA-3D 1,318 Elo for Astra vs 1,259 Elo for Sol; on Agents' Last Exam, Astra beats GPT-6 Astra by 5.2 points and Claude Fable 5.2 by 5.5 points; on ExploitBench Internal Port, Astra reaches 53.5% code execution (up 22 percentage points from 31.5% on GPT-6 Astra); the system card rates Cybersecurity as Critical and rates Astra as High on AI Self-Improvement (Sol does not reach High). ⑤ Artificial Analysis (2026-11-17): Intelligence Index 62 (#2) for Astra and 57 (#5) for Sol.

Not published or not supported by OpenAI

As of 2026-11-18, OpenAI has not published SWE-bench Verified scores for the 6.2 generation; evaluate them using the published FrontierCode 2.0, AA-3D, GPQA Diamond, and Agents' Last Exam figures instead. OpenAI also did not release any 6.2 mini, terra, luna, or dedicated codex variants; on endpoints, neither 6.2 model supports v1/realtime, v1/assistants, v1/fine-tuning, or predicted_outputs. Finally, OpenAI's Help Center has not yet announced the exact calendar day in December 2026 when the shared ChatGPT and Codex usage pool takes effect.

How to Choose by Engineering Workload

Everyday coding, CI pipelines & subagent fleets: pick gpt-6.2-sol

For day-to-day feature work, PR reviews, automated test generation, and high-concurrency subagent execution, gpt-6.2-sol delivers near-Astra coding accuracy (within 0.5 points on FrontierCode 2.0 at max) at one-fourth the OpenAI official unit price ($2 / $10 vs $8 / $40) with a newer 2026-06-01 knowledge cutoff. In Responses API Multi-agent beta workflows, routing worker subagents to gpt-6.2-sol keeps token spend predictable.

Hardest architecture refactors, deep reasoning & security audits: pick gpt-6.2-astra

For greenfield system architecture, tough concurrency bugs, long-horizon agent planning (+5.2 points over GPT-6 Astra on Agents' Last Exam), or security research (53.5% on ExploitBench Internal Port, 97.0% on GPQA Diamond, and #1 on AA-3D at 1,318 Elo), gpt-6.2-astra provides OpenAI's highest reasoning ceiling while cutting official unit prices by 20% ($8 / $40 vs $10 / $50 on gpt-6-astra).

Copy-Paste Setup for Codex CLI and OpenAI SDK on QCode

① Codex CLI (~/.codex/config.toml): set model_provider = "qcode", set model = "gpt-6.2-sol" for daily coding (or "gpt-6.2-astra" for your hardest tasks), and set model_reasoning_effort to "high" or "xhigh" (never set "none", which the API rejects); under [model_providers.qcode] set base_url = "https://api.qcode.cc/openai" (no /v1 suffix; use https://asia.qcode.cc/openai in mainland China or https://us.qcode.cc/openai in North America/Europe), wire_api = "responses", and requires_openai_auth = true, then put your cr_ key in OPENAI_API_KEY inside ~/.codex/auth.json. ② OpenAI SDK / Cline / WorkBuddy / OpenCode: set base_url to https://api.qcode.cc/openai/v1 (or https://asia.qcode.cc/openai/v1 in mainland China, with no trailing slash) and set model to gpt-6.2-sol or gpt-6.2-astra. For worked cost examples see /gpt-6-2-pricing, and for the dedicated Codex setup guide see /gpt-6-2-astra-codex.

Run Both 6.2 Models on One QCode Key

QCode supports both gpt-6.2-sol and gpt-6.2-astra, and the same cr_ API key also calls gpt-6.1-sol, gpt-6-astra, claude-fable-5-2, qwen4-max, and more. QCode bills per token, with live per-model rates on /models. Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; if you already have a Personal Plan, add a booster pack when you need more quota.

Frequently Asked Questions

How much do gpt-6.2-sol and gpt-6.2-astra cost on OpenAI's official API?

Per OpenAI's official API pricing (≤272K short-context Standard tier, per 1M tokens): gpt-6.2-sol is $2 input, $0.10 cached input, $2.50 30-minute cache write ($5 12-hour), and $10 output; gpt-6.2-astra is $8 input, $0.40 cached input, $10 30-minute cache write ($20 12-hour), and $40 output. Astra's standard rate is 4x Sol's, but 20% below gpt-6-astra ($10 / $50). QCode bills per token, with per-model rates on /models.

Why does setting reasoning_effort="none" fail on GPT-6.2?

OpenAI's official model pages state that both gpt-6.2-sol and gpt-6.2-astra reject API requests that set reasoning effort to "none". If an existing script or ~/.codex/config.toml sets reasoning_effort = "none", change it to "low", "medium", "high", "xhigh", or "max".

How does the >272K long-context pricing rule work?

Under OpenAI's official pricing rules, when a single request has more than 272K input tokens (up to the 922,000 max input limit), input and cache rates double (2x) and output costs 1.5x the short-context rate for the entire request: gpt-6.2-sol becomes $4 input / $15 output, and gpt-6.2-astra becomes $16 input / $60 output per 1M tokens. Keeping prompts at or below 272K and using prompt caching avoids the 2x/1.5x surcharge.

Should I upgrade from gpt-6.1-sol to gpt-6.2-sol?

Yes. gpt-6.2-sol keeps the exact same OpenAI official API unit prices as gpt-6.1-sol ($2 input / $0.10 cached input / $10 output) and the same 1,050,000 context window (922,000 max input / 128,000 max output), while advancing the knowledge cutoff from 2026-04-30 to 2026-06-01, adding Responses API Multi-agent beta support, and scoring within 0.5 points of gpt-6.2-astra on FrontierCode 2.0 at max.

Did OpenAI release a 6.2 mini, terra, or dedicated codex model ID?

No. As of 2026-11-18, the only official 6.2 API model IDs published by OpenAI are gpt-6.2-astra and gpt-6.2-sol; there are no 6.2 mini, terra, luna, or dedicated codex variants. Use gpt-6.2-sol or gpt-6.2-astra directly in Codex CLI.

Which QCode base_url should I use for Codex vs OpenAI SDK?

In Codex CLI (~/.codex/config.toml with wire_api = "responses"), set base_url to https://api.qcode.cc/openai (no /v1 suffix; https://asia.qcode.cc/openai in mainland China, https://us.qcode.cc/openai in North America/Europe). For the OpenAI SDK, Cline, WorkBuddy, or OpenCode using OpenAI-compatible chat/responses routes, set base_url to https://api.qcode.cc/openai/v1.

Sources

OpenAI official docs: GPT-6.2 Astra model page (developers.openai.com/api/docs/models/gpt-6.2-astra), GPT-6.2 Sol model page (developers.openai.com/api/docs/models/gpt-6.2-sol), OpenAI API pricing page, GPT-6.2 system card and release notes, and OpenAI Help Center article 20001625 (December 2026 shared ChatGPT and Codex usage pool); Artificial Analysis 2026-11-17 leaderboard; QCode: qcode.cc/models and docs.qcode.cc endpoint guides (all verified on 2026-11-18).

Call GPT-6.2 Sol and Astra with One QCode Key

Create a QCode account, billed per token with live per-model rates on /models; run gpt-6.2-sol for everyday coding and switch to gpt-6.2-astra for your hardest tasks.

Not sure where to start?

Try it for a week

Small Pack: $4 for $30 of quota, valid 7 days

Use it every day

Starter: $9.00/month with $5 of quota every day — just $0.060 per $1 of quota, vs about $0.113 for the Medium Pack