August 2025–August 2026

A year of model turnover

A reconstructed history of when the AI models in our survey arrived, when we first tested them, and how long each version remained in the research record.

24 versions surveyed7 providers24,040 successful runs5 current models

Release

The provider announcement or API-availability date, where one is public. Catalog-derived dates are labeled directly.

Surveyed

First and last successful Hallucination Yield run recorded during the 12-month window, in UTC.

Status

“Current survey” means the model is in today’s primary scheduled lineup. Historical models remain part of the dataset.

19 launches added

Release timeline

Newest first. Release dates and survey dates can differ because we validate availability and fit before adding a model to recurring research.

  1. •Provider announcement

    gemini-3.6-flash

    Google

    Current survey

    Survey coverage

    August 7, 2026–August 7, 2026

    Successful runs

    403

    Google launch announcement
  2. •Hallucination Yield catalog date

    grok-4.5

    xAI

    Current survey

    Survey coverage

    July 11, 2026–August 5, 2026

    Successful runs

    458

    xAI announcement on July 16
  3. •Provider announcement

    gpt-5.6-terra

    OpenAI

    Current survey

    Survey coverage

    July 11, 2026–August 5, 2026

    Successful runs

    458

    OpenAI launch announcement
  4. •Provider announcement

    claude-sonnet-5

    Anthropic

    Current survey

    Survey coverage

    July 11, 2026–August 5, 2026

    Successful runs

    457

    Anthropic launch announcement
  5. •Provider announcement

    gemini-3.5-flash

    Google

    Historical

    Survey coverage

    May 20, 2026–August 3, 2026

    Successful runs

    1,260

    Google launch announcement
  6. •Hallucination Yield catalog date

    grok-4.3

    xAI

    Historical

    Survey coverage

    May 8, 2026–July 11, 2026

    Successful runs

    923

  7. •API availability

    deepseek-v4-pro

    DeepSeek

    Current survey

    Survey coverage

    May 8, 2026–August 5, 2026

    Successful runs

    1,344

    DeepSeek API changelog
  8. •API availability

    gpt-5.5

    OpenAI

    Historical

    Survey coverage

    May 8, 2026–July 11, 2026

    Successful runs

    923

    OpenAI launch announcement
  9. •Provider announcement

    gpt-5.3-chat

    OpenAI

    Historical

    Survey coverage

    March 4, 2026–May 8, 2026

    Successful runs

    913

    OpenAI launch announcement
  10. •Provider announcement

    claude-sonnet-4.6

    Anthropic

    Historical

    Survey coverage

    March 4, 2026–July 11, 2026

    Successful runs

    1,435

    Anthropic launch announcement
  11. •Provider announcement

    gemini-3-flash-preview

    Google

    Historical

    Survey coverage

    January 16, 2026–May 20, 2026

    Successful runs

    1,783

    Google launch announcement
  12. •Provider announcement

    gpt-5.2-chat

    OpenAI

    Historical

    Survey coverage

    December 12, 2025–March 1, 2026

    Successful runs

    1,297

    OpenAI launch announcement
  13. •API availability

    deepseek-v3.2

    DeepSeek

    Historical

    Survey coverage

    December 12, 2025–May 6, 2026

    Successful runs

    1,809

    DeepSeek API changelog
  14. •API availability

    grok-4.1-fast

    xAI

    Historical

    Survey coverage

    January 16, 2026–May 8, 2026

    Successful runs

    1,564

    xAI launch announcement
  15. •Provider announcement

    kimi-k2-thinking

    Moonshot AI

    Historical

    Survey coverage

    November 8, 2025–November 8, 2025

    Successful runs

    68

    Moonshot AI technical post
  16. •API availability

    deepseek-v3.2-exp

    DeepSeek

    Historical

    Survey coverage

    September 29, 2025–December 10, 2025

    Successful runs

    898

    DeepSeek API changelog
  17. •Provider announcement

    claude-sonnet-4.5

    Anthropic

    Historical

    Survey coverage

    September 29, 2025–March 1, 2026

    Successful runs

    2,147

    Anthropic launch announcement
  18. •Provider announcement

    grok-4-fast

    xAI

    Historical

    Survey coverage

    September 29, 2025–January 11, 2026

    Successful runs

    1,336

    xAI launch announcement
  19. •API availability

    gpt-5-chat

    OpenAI

    Historical

    Survey coverage

    August 8, 2025–December 10, 2025

    Successful runs

    1,359

    OpenAI API announcement

Already in rotation

Models carried into the window

These versions launched earlier, but still produced successful survey data after August 6, 2025. Including them prevents the timeline from erasing the models that defined the starting point.

claude-sonnet-4

Anthropic

Carry-in
Released
May 22, 2025
Surveyed in window
August 6, 2025–October 2, 2025
Successful runs
483

Provider announcement

Anthropic launch announcement

deepseek-v3

DeepSeek

Carry-in
Released
March 24, 2025
Surveyed in window
August 6, 2025–October 2, 2025
Successful runs
444

API availability

DeepSeek API changelog

qwen-max

Alibaba Cloud

Carry-in
Released
January 25, 2025
Surveyed in window
August 6, 2025–October 2, 2025
Successful runs
484

Hallucination Yield catalog date

gpt-4o

OpenAI

Carry-in
Released
May 13, 2024
Surveyed in window
August 6, 2025–August 8, 2025
Successful runs
2

Provider announcement

OpenAI launch announcement

How to read this

A research ledger, not a benchmark

The timeline records model turnover in the Hallucination Yield survey. It does not claim that a newer model is better, or that every successful run produced the same number of observations.

“Successful runs” combines successful ranking, rating, and comparison jobs. Release dates are reconstructed from provider announcements and API changelogs, then matched to our exact surveyed model identifiers. When a public announcement could not be tied cleanly to the identifier, we retain and label the date from our internal model catalog.

Snapshot: August 7, 2026 UTC. Historical records may expand as earlier data is recovered or reprocessed.

See the current model lineupRead the methodology

Research only. Model outputs and this historical reconstruction may be incomplete or inaccurate. Nothing on this page is financial advice, and no warranty is provided.