Image support is unresolved: the 1 October prose announces Pro image input, but both 2024 capability tables show a cross for image analysis. Text is the uncontested modality.
No immutable archived selector verified. The modern GigaChat-Pro/GigaChat-Max aliases must not be treated as access to the 2024 weights.
Access
Historical API offering, currently unavailable according to provider release history. No active endpoint is asserted.
Regions
Historical processing region not verified.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Historical retention and training-use terms not collected.
Caching
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
GigaChat API (historical)
Not verified
No historical endpoint cache policy verified; current GigaChat terms are not applied retrospectively.
How it works
Not collected
No historical endpoint cache policy verified; current GigaChat terms are not applied retrospectively.
Minimum reusable input
Not collected
No historical endpoint cache policy verified; current GigaChat terms are not applied retrospectively.
Lifetime & refresh
Not collected
No historical endpoint cache policy verified; current GigaChat terms are not applied retrospectively.
How to enable it
Not collected
No historical endpoint cache policy verified; current GigaChat terms are not applied retrospectively.
Usage fields
Not collected
No historical endpoint cache policy verified; current GigaChat terms are not applied retrospectively.
Billing rules
Not collected
No historical endpoint cache policy verified; current GigaChat terms are not applied retrospectively.
Limits & data boundaries
Not collected
No historical endpoint cache policy verified; current GigaChat terms are not applied retrospectively.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 Oct 2026
What the measurements say
MMLU (5-shot) · 31.10.2024
Sber · provider-reported
0.69provider scale
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Arena-Hard EN (safety filter disabled) · 31.10.2024
Sber · provider-reported
18provider scale
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 31.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 01.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 01.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 01.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 01.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 01.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 01.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 01.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Historical 01.10.2024 release table. Raw published score with no rescaling; parenthesized change versus a prior release is not part of the score. Preserve the named few-shot and safety-filter settings. Not an independent reproduction or a current-model comparison.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
Official Sber model card compares the public GigaChat-20B-A3B checkpoint against separate GigaChat-Pro v26.20 and GigaChat-Max v26.20 models. The reviewed official public inventory contains no matching Pro/Max v26.20 weights. This classification does not extend to the public Lite model or newer GigaChat generations.
