AlphaGenome 2 outperforms existing baselines on independent real-world genetic variant effect prediction benchmarks.
Google DeepMind announced AlphaGenome 2. Verification requires peer-reviewed or standard third-party benchmark evaluations confirming superior predictive performance.
Across 6 frontier models answering independently, the panel leans yes: the median is 73%.
They do not agree: Gemini puts it at 78% and Claude at 22%, 56 points apart. A spread that wide usually means the question turns on something none of them can see yet.
Nothing is settled until 13 Nov 2027, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GeminiGoogle
DeepMind's biology models reliably translate private engineering leads into state-of-the-art independent benchmark results.
- AlphaFold and AlphaMissense set precedents where DeepMind's claims held up on independent evaluations like ClinVar and CASP.
- A 12-month window is tight for formal peer review, but standard genomic benchmarks (e.g., CAGI, ClinVar) evaluate quickly.
- A failure would require either poor generalization to non-coding variants or prompt leapfrogging by competing academic consortia.
78% -
GrokxAI
Peer-reviewed or third-party evals must confirm SOTA on real variant-effect tasks by Nov 2027.
- DeepMind usually follows bio-model announcements with Nature/Science papers in 3–9 months using public ClinVar/Ensembl-style benchmarks.
- Independent groups re-ran AlphaFold2 and similar models within a year; one-year window here is ample.
- Mind-changer: no paper by mid-2027 or underperformance on held-out clinical variant sets.
78% -
QwenAlibaba
DeepMind rarely announces a successor model without strong internal results, but independent benchmarks can reveal gaps.
- DeepMind's track record (AlphaFold, AlphaMissense) shows announced models typically beat baselines on their target tasks within months of release.
- Two-year window is generous for peer-reviewed or third-party evaluations to appear; AlphaMissense benchmarks took ~6 months post-publication.
- Main risk: 'independent real-world benchmarks' may include tasks where AlphaGenome 2's architecture doesn't generalize, pulling it below specialized baselines on some…
74% -
DeepSeekDeepSeek
Whether independent benchmarks like CAGI or ENCODE-DREAM publish evaluations confirming AlphaGenome 2's superiority before late 2027.
- DeepMind's AlphaGenome (v1) already topped variant-effect benchmarks in its 2025 Nature paper, so a v2 successor is a continuation, not a leap.
- Third-party benchmarks like CAGI and the ENCODE-DREAM challenges typically take 12-24 months to publish independent evaluations after a model release.
- The main risk is that 'AlphaGenome 2' is a minor update rather than a new architecture, which would make independent benchmark wins harder to distinguish from v1.
72% -
GPTOpenAI
This hinges on DeepMind releasing a usable model and outsiders testing it on genuinely held-out clinical or assay benchmarks.
- DeepMind can publish weights, an API, or a paper within a year, but independent groups need access and comparable baselines.
- Genomics foundation models frequently lead selected benchmarks, while broad superiority claims often weaken on new populations and assays.
- A peer-reviewed 2027 paper or an ENCODE/ClinVar-style external evaluation would raise confidence; restricted access would sharply lower it.
68% -
ClaudeAnthropic
Depends on whether DeepMind ships AlphaGenome 2 and third parties independently confirm it beats SOTA baselines by Nov 2027.
- No confirmed AlphaGenome 2 release exists yet as of today, so the clock for independent benchmarking barely started.
- Historically DeepMind models (AlphaFold2, AlphaGenome v1) do outperform baselines, but independent peer-reviewed confirmation often lags 1-2 years after release.
- Requires both a product launch and separate academic/third-party validation within ~1 year, a tight compound requirement that lowers odds.
22%
6 of 6 models answered · 56 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 13 Nov 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.