AI and science frontiers · 2026-10-29

Rare missense contributions to complex traits have been underestimated because collapsing variants within a gene into one burden score averages away effect differences. The Kernel Allelic Series Model (KASM) builds within-gene effect covariance kernels from protein-structure and variant-effect annotations; across 786,871 exomes and 26 continuous traits it raises the lower bound on rare missense heritability from 0.54±0.04% to at least 1.27±0.09%, and to 1.83±0.11% under the most flexible model, above the 0.87±0.05% for loss-of-function variants.

Rare missense contributions to complex traits have been underestimated because collapsing variants within a gene into one burden score averages away effect differences. The Kernel Allelic Series Model (KASM) builds within-gene effect covariance kernels from protein-structure and variant-effect annotations; across 786,871 exomes and 26 continuous traits it raises the lower bound on rare missense heritability from 0.54±0.04% to at least 1.27±0.09%, and to 1.83±0.11% under the most flexible model, above the 0.87±0.05% for loss-of-function variants.

The inverse-folding models used to design peptides emit a probability over exactly twenty natural residues at each position, and adding an unnatural one means supplying its structural data and retraining the whole model. AtomWeaver instead grows side chains as atoms in three dimensions under a fixed backbone and a target protein, then matches each finished atom cloud against a reference library of residue templates to read off identity. On PeptideArena, 353 backbones against twelve targets, it stayed competitive with twenty-residue methods and held the same performance regime with a 300-residue vocabulary.

The inverse-folding models used to design peptides emit a probability over exactly twenty natural residues at each position, and adding an unnatural one means supplying its structural data and retraining the whole model. AtomWeaver instead grows side chains as atoms in three dimensions under a fixed backbone and a target protein, then matches each finished atom cloud against a reference library of residue templates to read off identity. On PeptideArena, 353 backbones against twelve targets, it stayed competitive with twenty-residue methods and held the same performance regime with a 300-residue vocabulary.

Google released Gemini 4 Carbon, a next frontier model aimed at long-horizon agentic workflows and cybersecurity defense, ranking first on version 5.0 of the Artificial Analysis Intelligence Index at 59. It reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching, and says internal testing cut successful cheating on Terminal-Bench 4.0 to 0.0%. Access starts with a subset of cyber defenders and widens through the Fairwind Program, at the same price as Argon.

Google released Gemini 4 Carbon, a next frontier model aimed at long-horizon agentic workflows and cybersecurity defense, ranking first on version 5.0 of the Artificial Analysis Intelligence Index at 59. It reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching, and says internal testing cut successful cheating on Terminal-Bench 4.0 to 0.0%. Access starts with a subset of cyber defenders and widens through the Fairwind Program, at the same price as Argon.

Meta has released the next generation of Andromeda, its personalized ads retrieval system, designed jointly across model, system and GPU hardware. Retrieval previously could not afford ranking-stage model depth at its scale; the new architecture brings large-language-model-style mixture-of-experts into retrieval, approaching ranking-level performance at roughly 300x fewer inference FLOPs than the 2024 retrieval model. The combined single-GPU pipeline processes over one billion ads per second on one server, which Meta calls a new throughput benchmark for retrieval-stage models, and Meta reports a 2.4% conversion increase on Facebook since 2025.

Meta has released the next generation of Andromeda, its personalized ads retrieval system, designed jointly across model, system and GPU hardware. Retrieval previously could not afford ranking-stage model depth at its scale; the new architecture brings large-language-model-style mixture-of-experts into retrieval, approaching ranking-level performance at roughly 300x fewer inference FLOPs than the 2024 retrieval model. The combined single-GPU pipeline processes over one billion ads per second on one server, which Meta calls a new throughput benchmark for retrieval-stage models, and Meta reports a 2.4% conversion increase on Facebook since 2025.

AI data center siting, power procurement and cooling are still settled once by static single-point scoring on day-zero numbers, though water, power and carbon drift over a 10- to 20-year asset life and commitments bind for years. The build-out becomes a tiered Markov decision process: strategic commitments occur only in discrete windows, carry switching costs and take effect after source-dependent lead times, dispatch uses only capacity that exists, and a preference-conditioned multi-objective agent returns a policy family steerable at inference time across cost, carbon, water and reliability. On real site data, preference conditioning gives a confirmed, steerable trade-off on both objective pairs across three commitment regimes, and lock-in covers the frontier more evenly.

AI data center siting, power procurement and cooling are still settled once by static single-point scoring on day-zero numbers, though water, power and carbon drift over a 10- to 20-year asset life and commitments bind for years. The build-out becomes a tiered Markov decision process: strategic commitments occur only in discrete windows, carry switching costs and take effect after source-dependent lead times, dispatch uses only capacity that exists, and a preference-conditioned multi-objective agent returns a policy family steerable at inference time across cost, carbon, water and reliability. On real site data, preference conditioning gives a confirmed, steerable trade-off on both objective pairs across three commitment regimes, and lock-in covers the frontier more evenly.

PyTorch 2.15 makes the rewritten NCCL backend, previewed as opt-in nccl2 in 2.14, the default for every nccl process group, keeping TORCH_DIST_USE_NCCL2=0 and the nccl-legacy name as rollback paths. At cluster scale a single stalled GPU could halt a whole training run and the cause was often reproduced by hand; a backend-agnostic flight recorder now writes a post-mortem to disk when a collective hangs, and a timeout on any rank triggers dumps from every rank.

PyTorch 2.15 makes the rewritten NCCL backend, previewed as opt-in nccl2 in 2.14, the default for every nccl process group, keeping TORCH_DIST_USE_NCCL2=0 and the nccl-legacy name as rollback paths. At cluster scale a single stalled GPU could halt a whole training run and the cause was often reproduced by hand; a backend-agnostic flight recorder now writes a post-mortem to disk when a collective hangs, and a timeout on any rank triggers dumps from every rank.

Agent memory systems typically retrieve memories by relevance to the current query and use them as contextual evidence; TrawMem instead organizes interaction histories into local threads, then assigns task-oriented roles and jointly selects complementary memories to build a task workspace. Across LoCoMo and MemGallery with four open- and closed-source backbones, macro-F1 improves by 2.23 to 3.14 points over representative memory systems; on LoCoMo, LLM-Judge scores rise by 4.07 and 4.31 percentage points over SimpleMem on Qwen3.5-9B and DeepSeek-V4-Flash, with roughly 12.8% and 5.1% fewer context tokens.

Agent memory systems typically retrieve memories by relevance to the current query and use them as contextual evidence; TrawMem instead organizes interaction histories into local threads, then assigns task-oriented roles and jointly selects complementary memories to build a task workspace. Across LoCoMo and MemGallery with four open- and closed-source backbones, macro-F1 improves by 2.23 to 3.14 points over representative memory systems; on LoCoMo, LLM-Judge scores rise by 4.07 and 4.31 percentage points over SimpleMem on Qwen3.5-9B and DeepSeek-V4-Flash, with roughly 12.8% and 5.1% fewer context tokens.

Anthropic's life sciences team reports that Claude has been used inside real drug programs at six organizations, on tasks reaching from molecule design to manufacturing-document review and regulatory filings. Discussion of AI in drug development has centered on proposing initial molecule designs, while manufacturing paperwork stayed manual, with chemists spending up to 75% of their time on records. In Enveda's test, Claude caught all 11 issues human reviewers had flagged across 1,061 pages of documentation and flagged seven more, cutting human review time by 89%, from over 25 hours to less than three; most of these results remain preclinical.

Anthropic's life sciences team reports that Claude has been used inside real drug programs at six organizations, on tasks reaching from molecule design to manufacturing-document review and regulatory filings. Discussion of AI in drug development has centered on proposing initial molecule designs, while manufacturing paperwork stayed manual, with chemists spending up to 75% of their time on records. In Enveda's test, Claude caught all 11 issues human reviewers had flagged across 1,061 pages of documentation and flagged seven more, cutting human review time by 89%, from over 25 hours to less than three; most of these results remain preclinical.

On AMD Instinct MI355X, Qwen3-30B-A3B-Instruct trained with fully asynchronous agentic RL after supervised fine-tuning raised pass@1 from 7.3% to 27.5% on a 100-task stratified subset of SWE-bench Verified, with RL tasks drawn from repositories absent from that benchmark. The bottlenecks were long-tail trajectories and CPU sandbox scheduling rather than GPU compute: capping trajectory length cut step time from 850 s to 284 s, and cloud sandboxes at 384 concurrent trajectories lifted rollout throughput from about 860 to about 3,480 trajectories per hour.

On AMD Instinct MI355X, Qwen3-30B-A3B-Instruct trained with fully asynchronous agentic RL after supervised fine-tuning raised pass@1 from 7.3% to 27.5% on a 100-task stratified subset of SWE-bench Verified, with RL tasks drawn from repositories absent from that benchmark. The bottlenecks were long-tail trajectories and CPU sandbox scheduling rather than GPU compute: capping trajectory length cut step time from 850 s to 284 s, and cloud sandboxes at 384 concurrent trajectories lifted rollout throughput from about 860 to about 3,480 trajectories per hour.

Long-horizon language-model agent training now builds action preferences inside existing rollout groups, matching on pre-action state plus task progress, and raises mean WebShop success by 3.96 and 5.16 points on Qwen3-4B and 8B.

Long-horizon language-model agent training now builds action preferences inside existing rollout groups, matching on pre-action state plus task progress, and raises mean WebShop success by 3.96 and 5.16 points on Qwen3-4B and 8B.

Microsoft has made the Copilot harness built by GitHub engineers, already adopted company-wide, available through Microsoft Foundry so developers can build agents that reason, use tools and finish long-running tasks as a managed service. Previously, keeping up with frontier models meant building and maintaining your own harness or model router; now one harness reaches 20+ frontier models plus bring-your-own-key, with the platform routing work to the right model for the task. The harness runtime has been ported to Rust, which Microsoft says cuts session overhead and time-to-first token, and it already runs across product surfaces at the scale of millions of users.

Microsoft has made the Copilot harness built by GitHub engineers, already adopted company-wide, available through Microsoft Foundry so developers can build agents that reason, use tools and finish long-running tasks as a managed service. Previously, keeping up with frontier models meant building and maintaining your own harness or model router; now one harness reaches 20+ frontier models plus bring-your-own-key, with the platform routing work to the right model for the task. The harness runtime has been ported to Rust, which Microsoft says cuts session overhead and time-to-first token, and it already runs across product surfaces at the scale of millions of users.

ElevenLabs released Eleven Realtime Translation, a live voice translation model that lets each person speak their own language and hear the other in theirs, in a natural, expressive voice, while the conversation happens. Previously, a customer calling in one language often waited for an agent who understood them, or an interpreter had to join the call. The model runs on the low-latency speech model Eleven v4 Turbo, covers more than 90 languages with automatic language detection, keeps one voice consistent across languages, and is in alpha for ElevenAPI enterprise customers.

ElevenLabs released Eleven Realtime Translation, a live voice translation model that lets each person speak their own language and hear the other in theirs, in a natural, expressive voice, while the conversation happens. Previously, a customer calling in one language often waited for an agent who understood them, or an interpreter had to join the call. The model runs on the low-latency speech model Eleven v4 Turbo, covers more than 90 languages with automatic language detection, keeps one voice consistent across languages, and is in alpha for ElevenAPI enterprise customers.

Structured knowledge retrieval no longer needs predefined query-rule pairs: a routing agent picks a reasoning rule per query, its retriever ranks evidence under that rule's constraints, and the two train in alternation, each supervising the other.

Structured knowledge retrieval no longer needs predefined query-rule pairs: a routing agent picks a reasoning rule per query, its retriever ranks evidence under that rule's constraints, and the two train in alternation, each supervising the other.

Synchrotron X-ray diffraction produces a 2D diffractogram per second, and image models seem able to read α and β phase fractions of Ti–6Al–4V from them. Two models reach similar accuracy after fine-tuning on 250 experimental diffractograms, yet robustness tests and attention maps show the ImageNet-only model infers fractions from background and pixel noise, while the model pre-trained on simulated diffractograms relies on Bragg reflections. Accuracy alone does not validate a model.

Synchrotron X-ray diffraction produces a 2D diffractogram per second, and image models seem able to read α and β phase fractions of Ti–6Al–4V from them. Two models reach similar accuracy after fine-tuning on 250 experimental diffractograms, yet robustness tests and attention maps show the ImageNet-only model infers fractions from background and pixel noise, while the model pre-trained on simulated diffractograms relies on Bragg reflections. Accuracy alone does not validate a model.

Anthropic has put Claude models on Google Cloud in an Anthropic-managed form, so developers call the Claude API through their own Google Cloud project. Access is controlled by Google Cloud IAM and billing runs through Google Cloud Marketplace, replacing a separate arrangement for permissions and payment. Fast mode for Claude Opus 5.5 is available on the same managed channel.

Anthropic has put Claude models on Google Cloud in an Anthropic-managed form, so developers call the Claude API through their own Google Cloud project. Access is controlled by Google Cloud IAM and billing runs through Google Cloud Marketplace, replacing a separate arrangement for permissions and payment. Fast mode for Claude Opus 5.5 is available on the same managed channel.

A preference optimization reward built on the psycholinguistic uniform information density hypothesis matches SimPO on AlpacaEval 2, Arena-Hard and MT-Bench, and moves English complement-clause that usage closer to natural language.

A preference optimization reward built on the psycholinguistic uniform information density hypothesis matches SimPO on AlpacaEval 2, Arena-Hard and MT-Bench, and moves English complement-clause that usage closer to natural language.

Vercel introduced Virtual Models on AI Gateway, so apps and agents call a task-named virtual model ID instead of a concrete model name, with model choice, fallbacks and Zero Data Retention configured on the gateway. Previously switching models was a one-line code change that still usually required a redeploy; now editing the virtual model is picked up by every app and agent using that ID within a few minutes.

Vercel introduced Virtual Models on AI Gateway, so apps and agents call a task-named virtual model ID instead of a concrete model name, with model choice, fallbacks and Zero Data Retention configured on the gateway. Previously switching models was a one-line code change that still usually required a redeploy; now editing the virtual model is picked up by every app and agent using that ID within a few minutes.

CopilotKit released OpenBot, a cloneable self-hosted template for an AI agent workspace in which each coworker agent gets its own browser, logins and files and only the tools it is granted. Any agent speaking the AG-UI protocol joins the same way, and every action on a computer, file, MCP server or component passes through one gateway that decides before and records after. A deployment starts with two configuration-defined coworker agents and ten more fintech examples; the project is in beta, with no hosted version and no published package.

CopilotKit released OpenBot, a cloneable self-hosted template for an AI agent workspace in which each coworker agent gets its own browser, logins and files and only the tools it is granted. Any agent speaking the AG-UI protocol joins the same way, and every action on a computer, file, MCP server or component passes through one gateway that decides before and records after. A deployment starts with two configuration-defined coworker agents and ten more fintech examples; the project is in beta, with no hosted version and no published package.

In multimodal long-document question answering, agentic retrieval is not always worth its extra computation and can even hurt answer accuracy. RouteDRAG makes both whether to invoke an agent and how it then executes into learned choices: a graph-backed index organizes cross-page, table and figure content, path routing predicts the benefit of an agent over shorter paths, and dynamic execution controls tool use and termination. Across two benchmarks it assigns only 79.3% of queries to an agent while averaging 3.47 percentage points higher accuracy than shorter-path-only execution, and it also surpasses uniform agent execution with fewer answering calls.

In multimodal long-document question answering, agentic retrieval is not always worth its extra computation and can even hurt answer accuracy. RouteDRAG makes both whether to invoke an agent and how it then executes into learned choices: a graph-backed index organizes cross-page, table and figure content, path routing predicts the benefit of an agent over shorter paths, and dynamic execution controls tool use and termination. Across two benchmarks it assigns only 79.3% of queries to an agent while averaging 3.47 percentage points higher accuracy than shorter-path-only execution, and it also surpasses uniform agent execution with fewer answering calls.

Baseten's research lab released two open-source decision models, mammoth-1 (27B) and elephant-1 (9B), which return a probability distribution over every possible answer instead of generating text. mammoth-1 ranks first among 208 decision models on the Jev Decision Index 0.4, while elephant-1 is tied for eighth there and is the leading 9B decision model, and is tied for first with d1-ultra on the separate Artificial Analysis Decision Index. Both were post-trained from Qwen3.5-9B and Qwen3.5-27B on 6.21 million decision-making examples, are free to try for a week on Baseten Model APIs, and are open-source on Hugging Face.

Baseten's research lab released two open-source decision models, mammoth-1 (27B) and elephant-1 (9B), which return a probability distribution over every possible answer instead of generating text. mammoth-1 ranks first among 208 decision models on the Jev Decision Index 0.4, while elephant-1 is tied for eighth there and is the leading 9B decision model, and is tied for first with d1-ultra on the separate Artificial Analysis Decision Index. Both were post-trained from Qwen3.5-9B and Qwen3.5-27B on 6.21 million decision-making examples, are free to try for a week on Baseten Model APIs, and are open-source on Hugging Face.

Gaze control for a humanoid face robot is split in two: a vision-language model asynchronously picks the scene referent, and a Realistic Gaze Transformer turns a target eye trajectory into coordinated neck, binocular-eye, and eyelid motion. The model is a Transformer-VAE trained on markerless human recordings, first self-supervised and then autoregressively fine-tuned so training includes its own predictions. On a two-second closed-loop benchmark where every controller receives the same ground-truth eye target, RGT-Base reaches the lowest reported seven-joint error, with inference inside the 16.7 ms frame interval.

Gaze control for a humanoid face robot is split in two: a vision-language model asynchronously picks the scene referent, and a Realistic Gaze Transformer turns a target eye trajectory into coordinated neck, binocular-eye, and eyelid motion. The model is a Transformer-VAE trained on markerless human recordings, first self-supervised and then autoregressively fine-tuned so training includes its own predictions. On a two-second closed-loop benchmark where every controller receives the same ground-truth eye target, RGT-Base reaches the lowest reported seven-joint error, with inference inside the 16.7 ms frame interval.

Cloudflare added internal hostnames to the persistent browser in its Agents SDK: give the browser a name such as preview.internal and requests to it go to the agent itself rather than the internet. The agent can then check a dev server in its sandbox, a folder in its workspace, or another Worker in the same browser it uses for the web, with nothing exposed publicly. Each internal host has a fetch function running in the Durable Object, so it can use that object's bindings and state, and a host can carry a target schema so the model chooses which port or folder it serves.

Cloudflare added internal hostnames to the persistent browser in its Agents SDK: give the browser a name such as preview.internal and requests to it go to the agent itself rather than the internet. The agent can then check a dev server in its sandbox, a folder in its workspace, or another Worker in the same browser it uses for the web, with nothing exposed publicly. Each internal host has a fetch function running in the Durable Object, so it can use that object's bindings and state, and a host can carry a target schema so the model chooses which port or folder it serves.

Baseten now lets read-only model data be published once as a sealed volume: a directory or checkpoint is committed as an immutable version addressed by a content digest that any deployment can mount. Previously the same checkpoint was downloaded and re-uploaded to several places, each deployment holding its own config pointing at a copy, so a mismatched revision surfaced only when outputs diverged. Volumes are chunked and keyed by content hash, so a push transfers only changed chunks, and a mount makes files available as the container starts, removing the pull phase from startup. Sealed Volumes ship with Baseten CLI 1.1.0 and newer.

Baseten now lets read-only model data be published once as a sealed volume: a directory or checkpoint is committed as an immutable version addressed by a content digest that any deployment can mount. Previously the same checkpoint was downloaded and re-uploaded to several places, each deployment holding its own config pointing at a copy, so a mismatched revision surfaced only when outputs diverged. Volumes are chunked and keyed by content hash, so a push transfers only changed chunks, and a mount makes files available as the container starts, removing the pull phase from startup. Sealed Volumes ship with Baseten CLI 1.1.0 and newer.

A memory layer for agents doing cross-document reasoning now has a controlled comparison: on 111 synthetic insurance documents, an agent using the OpenWiki-compiled wiki reached 0.52 accuracy on 50 questions requiring connections across sources, against 0.10 with keyword search alone; adding an evidence graph built from citations and document references raised accuracy a further 27%, doubled citation consistency and cut cost per correct answer by 58%. On simple lookups the two conditions were about equal, so the wiki brought no advantage there.

A memory layer for agents doing cross-document reasoning now has a controlled comparison: on 111 synthetic insurance documents, an agent using the OpenWiki-compiled wiki reached 0.52 accuracy on 50 questions requiring connections across sources, against 0.10 with keyword search alone; adding an evidence graph built from citations and document references raised accuracy a further 27%, doubled citation consistency and cut cost per correct answer by 58%. On simple lookups the two conditions were about equal, so the wiki brought no advantage there.

Luminary released SHIFT-NVH, a large physics model trained on finite-element solves of a single body-in-white that takes the gauges of 15 parts, the material and the vehicle speed and returns RMS acceleration in x, y and z at all 130,713 surface nodes under ISO 8608 class B road excitation. On 192 test cases split by design, with material and speed unseen, the median relative L2 field error is 1.1% and the median part RMS error is 0.47%, about one tenth of the 5.4% median gauge-induced change. The model covers a fixed mesh, ±30% gauge, three materials and four speeds from 30 to 110 km/h, and final gauge decisions still rest on full FE and physical testing.

Luminary released SHIFT-NVH, a large physics model trained on finite-element solves of a single body-in-white that takes the gauges of 15 parts, the material and the vehicle speed and returns RMS acceleration in x, y and z at all 130,713 surface nodes under ISO 8608 class B road excitation. On 192 test cases split by design, with material and speed unseen, the median relative L2 field error is 1.1% and the median part RMS error is 0.47%, about one tenth of the 5.4% median gauge-induced change. The model covers a fixed mesh, ±30% gauge, three materials and four speeds from 30 to 110 km/h, and final gauge decisions still rest on full FE and physical testing.

AI Gateway now offers virtual models: a saved configuration naming which model to use and with which settings, called by apps and agents through a custom model ID such as vmc/coding-agent. Previously, changing the model or settings meant redeploying apps or reconfiguring agents; now the change happens centrally in AI Gateway. It can also supply options a coding agent does not expose, such as reasoning effort, fast mode and prompt caching, and carry provider choice, fallback models and Zero Data Retention requirements.

AI Gateway now offers virtual models: a saved configuration naming which model to use and with which settings, called by apps and agents through a custom model ID such as vmc/coding-agent. Previously, changing the model or settings meant redeploying apps or reconfiguring agents; now the change happens centrally in AI Gateway. It can also supply options a coding agent does not expose, such as reasoning effort, fast mode and prompt caching, and carry provider choice, fallback models and Zero Data Retention requirements.

Anthropic has made three admin features generally available for Claude Enterprise: smart reports sample a team's transcripts, group the work into workstreams and show each one's spend, deliverables and common friction points; in-app surveys collect members' own account of value; and pooled group budgets give a team one shared monthly budget while member limits remain a guardrail. All three shipped the same day, with availability varying by configuration, including HIPAA configurations and organizations using customer-managed encryption keys.

Anthropic has made three admin features generally available for Claude Enterprise: smart reports sample a team's transcripts, group the work into workstreams and show each one's spend, deliverables and common friction points; in-app surveys collect members' own account of value; and pooled group budgets give a team one shared monthly budget while member limits remain a guardrail. All three shipped the same day, with availability varying by configuration, including HIPAA configurations and organizations using customer-managed encryption keys.

The Gemini app now carries a library of immersive explorations that lets users move through a topic layer by layer in an interactive, visual form, from microscopic systems out to deep space. The library is reachable in the Gemini student hub through “Try immersive explorations,” and direct access inside a Gemini chat is stated as coming soon. Each scene unfolds into guided, interactive experiences, such as tracing trade routes on the Silk Road, peering inside a jet engine, or studying the physics of lightning. The feature is available to all users in English, Indonesian, Hindi, Brazilian Portuguese, French, and simplified Chinese, while users in all other languages currently receive an English-only experience.

The Gemini app now carries a library of immersive explorations that lets users move through a topic layer by layer in an interactive, visual form, from microscopic systems out to deep space. The library is reachable in the Gemini student hub through “Try immersive explorations,” and direct access inside a Gemini chat is stated as coming soon. Each scene unfolds into guided, interactive experiences, such as tracing trade routes on the Silk Road, peering inside a jet engine, or studying the physics of lightning. The feature is available to all users in English, Indonesian, Hindi, Brazilian Portuguese, French, and simplified Chinese, while users in all other languages currently receive an English-only experience.

Selecting five models by cognitive diversity from a pool of 29 quantum and classical scoring systems and fusing them beats the strongest single model on all four binary classification benchmarks, with gains up to 6.03% in accuracy.

Selecting five models by cognitive diversity from a pool of 29 quantum and classical scoring systems and fusing them beats the strongest single model on all four binary classification benchmarks, with gains up to 6.03% in accuracy.