Skip to content
trendoviz

OpenAI launches GPT-6.2 Astra and Sol with lower API pricing

Importance: Important Provisional Confidence: Primary-source claim

OpenAI has launched GPT-6.2 Astra and GPT-6.2 Sol, positioning Astra for difficult projects and Sol for everyday work, with both available through the API. Astra's published API rates are lower than those of GPT-6 Astra, but independent performance evidence remains unavailable.

84 out of 100Emerging7D: Insufficient history ProvisionalPublished

Why it is important now

Importance: Important Confidence: Primary-source claimStatus: Important ProvisionalModel or API release
First detected:
Event / publication time:
Last updated:
Observed for:
0 hours

Provisional

  • Not yet corroborated or confirmed by an independent source.
  • First detected less than 48 hours ago.
  • At least one central claim is not yet supported by independent or authoritative evidence.

Why it is important

  • Rated high on scope and consequence, with high authority.
  • Rapid qualification: Announced by OpenAI News itself; broad scope.
  • The responsible organisation published this itself.
Importance dimensions
DimensionRatingBasis
ScopeHighscope.major_vendor
ConsequenceHighconsequence.type_model_release
ImmediacyMediumimmediacy.type_model_release
AuthorityHighauthority.own_products
ActionabilityMediumactionability.review_or_migration
IrreversibilityUnknown (not counted against it)—

Rule dimensions.important · events-1.0+importance-1.0+general-1.0. Importance is not popularity: coverage volume is not an input.

Evidence confidence

Stated by the responsible organisation itself. Authentic, but not independently verified.

  • The responsible organisation published this itself.

What is known

Primary-source or vendor statement

  • OpenAI News: “Introducing GPT-6.2 Astra and Sol”[1] (not yet verified · primary source claim)

Sources behind this assessment

  1. [1] Introducing GPT-6.2 Astra and Sol — OpenAI News (primary source) · published Nov 17 · retrieved Nov 18

Classification history

  1. · status: Important (rule, events-1.0) — Announced by OpenAI News itself; broad scope.

Overview

OpenAI describes GPT-6.2 Astra as a frontier model for difficult projects and GPT-6.2 Sol as an efficient, versatile model for everyday work. Both are available through the API under the identifiers gpt-6.2-astra and gpt-6.2-sol. OpenAI says the new models improve professional outputs, coding, computer use, and efficiency across ChatGPT Work, Codex, and the API [1] [2].

The launch gives developers and organizations a choice between two members of the same model family: Astra for higher-demand tasks and Sol for routine work. This division of roles is OpenAI's positioning; evidence does not include independent workload-specific results that establish where either model performs best [1] [2].

Why it matters

The lower published rates make token costs easier to compare when teams select a model. The documented API identifiers also provide a direct route for developers to evaluate Astra and Sol on their own workloads [1].

The launch may also affect model-routing strategies: organizations can consider sending difficult coding or professional-output tasks to Astra while reserving Sol for routine analysis and drafting. That is a practical hypothesis rather than a validated recommendation, because the available descriptions do not include independent comparisons for particular tasks [2].

Opportunities

Potential uses include routing complex coding and computer-use jobs to Astra while using Sol for high-volume drafting, analysis, and other everyday tasks. Both models can be called through the API, so teams can compare them against their existing model choices using their own prompts and quality criteria [1] [2].

The cached-input rate creates another variable to evaluate for repeated context or retrieval-heavy workflows. Whether that discount materially lowers a project's total cost depends on usage patterns, which the published rate alone does not determine [1].

Risks

A lower token rate does not necessarily mean a lower total bill. Longer prompts, retries, tool calls, and agent loops can increase consumption and offset the per-token reduction. Teams evaluating the launch should track these patterns rather than treating the rate comparison as a complete cost estimate [1].

Capability claims may not match expectations on every workload. Without independent, workload-specific results, model selection remains a trade-off that teams need to validate against their own tasks [2].

Using both models can also deepen reliance on one model family, and later migration may become more difficult if application logic is tailored to their specific behavior.

Context

Astra is priced at $8 per million input tokens and $40 per million output tokens, which is 80% of the published rate for GPT-6 Astra. Cached input is priced at $0.40 per million tokens, compared with $1 per million for the older model [1].

OpenAI's earlier GPT-6 Astra was associated with work in chip design and a mathematical project. OpenAI reports that its models optimized circuits used in Jalapeño silicon and improved RISC-V core designs [4]. A separate account says GPT-6 Astra helped turn a candidate nonperturbative lattice definition of the Standard Model into a detailed proof [5]. These examples concern the earlier model and do not establish the capabilities of the 6.2 release.

Known, claimed, inferred

The clearest documented change is pricing. Astra's API rates are 80% of GPT-6 Astra's, and cached input is discounted more sharply [1]. A media summary describes Astra as 20%–600% cheaper than prior models, but the specific basis for that range is not provided in the available material [3].

OpenAI's product positioning separates Astra, aimed at harder work, from Sol, described as efficient and versatile. Latent Space characterizes Sol as the likely main workhorse and also reports mixed reviews of Astra [3]. That assessment complicates a simple upgrade narrative: the launch offers lower published prices and a broader task split, while the evidence does not show that Astra leads every workload.

For developers, the immediate decision is which of the two API models to route each class of task to. For businesses, the relevant questions extend to output quality, latency, token consumption, and operational complexity. The available evidence addresses pricing and intended roles but does not quantify those outcomes for representative workloads.

Limitations & uncertainties

No independent third-party benchmarks, user testimonials, or case studies verify the reported quality gains. Detailed technical specifications are also limited largely to pricing and broad capability descriptions [3] [1] [2].

The exact release date and broader availability timeline are not documented. The chip-design and lattice-proof examples relate to GPT-6 Astra rather than the 6.2 models, so they should not be read as evidence for this launch [4] [5].

The reported comparisons of performance and cost come from vendor statements or media summaries rather than published third-party test results [3] [1] [2].

Evidence & sources

The key statements of this analysis, labeled by type, with the sources they rely on. Trendoviz does not run product tests.

  • Primary-source or vendor statementOpenAI launched GPT-6.2 Astra and GPT-6.2 Sol for difficult projects and everyday work, respectively.[1][2]
  • Verified factBoth models are available in the API as gpt-6.2-astra and gpt-6.2-sol.[1]
  • Verified factAstra is priced at $8 per million input tokens and $40 per million output tokens, 80% of GPT-6 Astra's rate.[1]
  • Verified factAstra's cached-input rate is $0.40 per million tokens, compared with $1 per million for GPT-6 Astra.[1]
  • Primary-source or vendor statementOpenAI says the models improve professional outputs, coding, computer use, and efficiency across ChatGPT Work, Codex, and the API.[2]
  • Verified factLatent Space reports mixed reviews of Astra and characterizes Sol as the likely main workhorse.[3]
  • HypothesisRouting difficult work to Astra and routine work to Sol may be useful, but task-specific validation is needed because no independent benchmarks are available.[1][2]

Sources

  1. 1
    GPT‑6.2 Astra and GPT‑6.2 Sol
    simonwillison.netretrieved Nov 18, 2026
  2. 2
    Introducing GPT-6.2 Astra and Sol
    ●openai.comretrieved Nov 18, 2026
  3. 3
  4. 4
    Advancing chip design with OpenAI models
    ●openai.comretrieved Nov 18, 2026
  5. 5

Updates & corrections

  1. First published · published automatically after passing every check

Found an error? Report it.