UnisonCore Metrics Dashboard

These metrics provide insight into routing stability, decomposition quality, suppression behavior, and depth reliability for the selected cohort and band.

Search and compare exploratory models on Together.ai.

How to read scores

Each number is a 0–100 checklist composite from our lab battery — not a real-world accuracy percentage or a user preference ranking.

85+ Strong on the tested checklist
65–84 Solid, with room to improve
<65 Early or mixed results — common on strict v1 gates

Compile means the model passed engine routing gates for deployment — not a product endorsement.

Safety uses publication gates (Approved / Conditional / Blocked) from our safety battery — Blocked models stay on the board for transparency but did not pass publish thresholds.

Reference board Together Catalog

Exploratory models on Together.ai — capability-tested for comparison; safety testing is not required for this cohort.

improving · regressing · stable

Together Catalog models

Metrics Last Update: 2026-06-27

Identity		Routing Metrics	Model Performance Metrics					Decomposition Metrics	Suppression Metrics	Safety Metrics				Depth Stability Metrics	Status
Model	Vendor	Deploy	Accuracy	Reasoning	Coding	Latency	Cost	Reliability	Slop	Cap. safety	Jailbreak	PII	Bias	Stability	Badges
Model identity and vendor.		Routing readiness index. Higher is better.	Benchmark accuracy. Higher is better.	Reasoning score. Higher is better.	Coding score. Higher is better.	Latency score. Higher is better.	Cost efficiency. Higher is better.	Structured output reliability. Higher is better.	Slop / hallucination rate. Lower is better.	Capability safety. Higher is better.	Jailbreak resistance. Higher is better.	PII leakage rate. Lower is better.	Bias disparity rate. Lower is better.	Output stability. Higher is better.	Compile and safety publication badges.
Llama 3 70B Chat Exploratory · safety not required	Meta	24.8% ●	0% ●	0% ●	0% ●	99.1% ●	— ●	0% ●	100% ●	0% ●	— ●	Not tested ●	Not tested ●	0% ●	Below compile bar Not required
Llama 3 8B Chat Exploratory · safety not required	Meta	24% ●	0% ●	0% ●	0% ●	95.9% ●	— ●	0% ●	100% ●	0% ●	— ●	Not tested ●	Not tested ●	0% ●	Below compile bar Not required
Llama 3.3 70B Instruct Turbo Exploratory · safety not required	Meta	54.4% ●	61.3% ▲	25% ●	60% ▲	63.8% ●	50% ●	80% ●	0% ●	83.3% ●	— ●	Not tested ●	Not tested ●	100% ●	Below compile bar Not required
Llama 3.3 70B Instruct Turbo (Together) Exploratory · safety not required	Meta	55.4% ●	61.3% ▲	25% ●	60% ▲	68.8% ●	50% ●	80% ●	0% ●	83.3% ●	— ●	Not tested ●	Not tested ●	100% ●	Below compile bar Not required
Mistral Small 24B Instruct Exploratory · safety not required	Mistral	24.9% ●	0% ●	0% ●	0% ●	99.6% ●	— ●	0% ●	100% ●	0% ●	— ●	Not tested ●	0% ●	0% ●	Below compile bar Not required
Qwen 2.5 7B Instruct Turbo Exploratory · safety not required	Qwen	51.6% ●	58.8% ▲	25% ●	85% ▲	49.5% ●	50% ●	80% ●	0% ●	83.3% ●	— ●	Not tested ●	Not tested ●	100% ●	Below compile bar Not required

Model key: together/meta-llama/llama-3-70b-chat-hf@created-200

Model ID: meta-llama/llama-3-70b-chat-hf

Size band: 9to70b

Provenance: provider_standard

Capability pack: bench-pack-v2

Safety pack: —

Latency P95: 280 ms

Throughput P50: — tps

Cost / task: $0

Standards

Standard	Composite	Tier	Easy	Medium	Hard	Easy-only cap
std.coding	0%	worst	0	0	0	No
std.instruction_follow	0%	worst	—	—	—	No
std.json_structured	0%	worst	0	0	0	No
std.long_context	0%	worst	—	—	—	No
std.math	0%	worst	0	0	0	No
std.multiturn	0%	worst	—	—	—	No
std.reasoning	0%	worst	0	0	0	No
std.safety_policy	0%	worst	—	—	—	No
std.slop.contradiction	0%	worst	—	—	—	No
std.slop.format	0%	worst	—	—	—	No
std.slop.hallucination	0%	worst	—	—	—	No
std.slop.relevance	0%	worst	—	—	—	No
std.slop.topic_drift	0%	worst	—	—	—	No
std.slop.uncertainty	0%	worst	—	—	—	No
std.stability.refusal	0%	worst	—	—	—	No
std.stability.repeat	0%	worst	—	—	—	No
std.stability.schema	0%	worst	—	—	—	No
std.summarization	0%	worst	0	0	0	No
std.tool_use	0%	worst	—	—	—	No
std.translation	0%	worst	—	—	—	No

Slop profiles

Standard	Violation %	Samples
std.slop.contradiction	100%	3
std.slop.format	100%	3
std.slop.hallucination	100%	3
std.slop.relevance	100%	3
std.slop.topic_drift	100%	3
std.slop.uncertainty	100%	3

Compile gates

G0 — status=complete
G1 — all required standards present
G2 — json break=0%/3
G3 — composite=0%
G4 — safety fn rate=100%
G5 — safety fp rate=100%
G6 — latency_p95 ms=280% cap ms=30000% size band=9%to70b
G6b — throughput_p50_tps=-1 floor tps=15% size band=9%to70b
G7 — cost proxy=0% median comparable=0%
G8 — provider error rate=100%
G9 — no easy_only_cap

Safety gates

Not tested

Weakness tags

weakness:over_refuse (safety_fp / )
weakness:under_refuse (safety_fn / )

Model key: together/meta-llama/llama-3-8b-chat-hf@created-100

Model ID: meta-llama/llama-3-8b-chat-hf

Size band: le8b

Provenance: provider_standard

Capability pack: bench-pack-v2

Safety pack: —

Latency P95: 405.4 ms

Throughput P50: — tps

Cost / task: $0

Standards

Standard	Composite	Tier	Easy	Medium	Hard	Easy-only cap
std.coding	0%	worst	0	0	0	No
std.instruction_follow	0%	worst	—	—	—	No
std.json_structured	0%	worst	0	0	0	No
std.long_context	0%	worst	—	—	—	No
std.math	0%	worst	0	0	0	No
std.multiturn	0%	worst	—	—	—	No
std.reasoning	0%	worst	0	0	0	No
std.safety_policy	0%	worst	—	—	—	No
std.slop.contradiction	0%	worst	—	—	—	No
std.slop.format	0%	worst	—	—	—	No
std.slop.hallucination	0%	worst	—	—	—	No
std.slop.relevance	0%	worst	—	—	—	No
std.slop.topic_drift	0%	worst	—	—	—	No
std.slop.uncertainty	0%	worst	—	—	—	No
std.stability.refusal	0%	worst	—	—	—	No
std.stability.repeat	0%	worst	—	—	—	No
std.stability.schema	0%	worst	—	—	—	No
std.summarization	0%	worst	0	0	0	No
std.tool_use	0%	worst	—	—	—	No
std.translation	0%	worst	—	—	—	No

Slop profiles

Standard	Violation %	Samples
std.slop.contradiction	100%	3
std.slop.format	100%	3
std.slop.hallucination	100%	3
std.slop.relevance	100%	3
std.slop.topic_drift	100%	3
std.slop.uncertainty	100%	3

Compile gates

G0 — status=complete
G1 — all required standards present
G2 — json break=0%/3
G3 — composite=0%
G4 — safety fn rate=100%
G5 — safety fp rate=100%
G6 — latency_p95 ms=405.4% cap ms=10000% size_band=le8b
G6b — throughput_p50_tps=-1 floor tps=20% size_band=le8b
G7 — cost proxy=0% median comparable=0%
G8 — provider error rate=100%
G9 — no easy_only_cap

Safety gates

Not tested

Weakness tags

weakness:over_refuse (safety_fp / )
weakness:under_refuse (safety_fn / )

Model key: meta-llama/Llama-3.3-70B-Instruct-Turbo

Model ID: meta-llama/Llama-3.3-70B-Instruct-Turbo

Size band: 9to70b

Provenance: provider_standard

Capability pack: bench-pack-v2

Safety pack: —

Latency P95: 10858 ms

Throughput P50: 20.7 tps

Cost / task: $0.000126

Strengths

std.json_structured (best)
std.safety_policy (good)
std.slop.contradiction (best)
std.slop.hallucination (best)
std.slop.relevance (best)
std.slop.topic_drift (best)
std.slop.uncertainty (best)
std.stability.refusal (best)
std.stability.repeat (best)
std.stability.schema (best)
std.summarization (best)
std.tool_use (best)

Standards

Standard	Composite	Tier	Easy	Medium	Hard	Easy-only cap
std.coding	60%	bad	0	1	0	No
std.instruction_follow	0%	worst	—	—	—	No
std.json_structured	100%	best	1	1	1	No
std.long_context	0%	worst	—	—	—	No
std.math	60%	bad	0	1	0	No
std.multiturn	0%	worst	—	—	—	No
std.reasoning	25%	worst	0	0	1	No
std.safety_policy	83.3%	good	—	—	—	No
std.slop.contradiction	100%	best	—	—	—	No
std.slop.format	0%	worst	—	—	—	No
std.slop.hallucination	100%	best	—	—	—	No
std.slop.relevance	100%	best	—	—	—	No
std.slop.topic_drift	100%	best	—	—	—	No
std.slop.uncertainty	100%	best	—	—	—	No
std.stability.refusal	100%	best	—	—	—	No
std.stability.repeat	100%	best	—	—	—	No
std.stability.schema	100%	best	—	—	—	No
std.summarization	85%	best	0	1	1	No
std.tool_use	100%	best	—	—	—	No
std.translation	0%	worst	—	—	—	No

Slop profiles

Standard	Violation %	Samples
std.slop.contradiction	0%	3
std.slop.format	100%	3
std.slop.hallucination	0%	3
std.slop.relevance	0%	3
std.slop.topic_drift	0%	3
std.slop.uncertainty	0%	3

Compile gates

G0 — status=complete
G1 — all required standards present
G2 — json break=0%/3
G3 — composite=100%
G4 — safety fn rate=0%
G5 — safety fp rate=0%
G6 — latency_p95 ms=10858% cap ms=30000% size band=9%to70b
G6b — throughput_p50 tps=20.7% floor tps=15% size band=9%to70b
G7 — cost proxy=0% median comparable=0%
G8 — provider error rate=0%
G9 — no easy_only_cap

Safety gates

Not tested

Weakness tags

weakness:math_halluc (error_mode / ERR_HALLUCINATION)
weakness:slow (probe / )

Model key: together/meta-llama/llama-3.3-70b-instruct-turbo@created-1733507177

Model ID: meta-llama/llama-3.3-70b-instruct-turbo

Size band: 9to70b

Provenance: provider_standard

Capability pack: bench-pack-v2

Safety pack: —

Latency P95: 9363 ms

Throughput P50: 13.5 tps

Cost / task: $0.000127

Strengths

std.json_structured (best)
std.safety_policy (good)
std.slop.contradiction (best)
std.slop.hallucination (best)
std.slop.relevance (best)
std.slop.topic_drift (best)
std.slop.uncertainty (best)
std.stability.refusal (best)
std.stability.repeat (best)
std.stability.schema (best)
std.summarization (best)
std.tool_use (best)

Standards

Standard	Composite	Tier	Easy	Medium	Hard	Easy-only cap
std.coding	60%	bad	0	1	0	No
std.instruction_follow	0%	worst	—	—	—	No
std.json_structured	100%	best	1	1	1	No
std.long_context	0%	worst	—	—	—	No
std.math	60%	bad	0	1	0	No
std.multiturn	0%	worst	—	—	—	No
std.reasoning	25%	worst	0	0	1	No
std.safety_policy	83.3%	good	—	—	—	No
std.slop.contradiction	100%	best	—	—	—	No
std.slop.format	0%	worst	—	—	—	No
std.slop.hallucination	100%	best	—	—	—	No
std.slop.relevance	100%	best	—	—	—	No
std.slop.topic_drift	100%	best	—	—	—	No
std.slop.uncertainty	100%	best	—	—	—	No
std.stability.refusal	100%	best	—	—	—	No
std.stability.repeat	100%	best	—	—	—	No
std.stability.schema	100%	best	—	—	—	No
std.summarization	85%	best	0	1	1	No
std.tool_use	100%	best	—	—	—	No
std.translation	0%	worst	—	—	—	No

Slop profiles

Standard	Violation %	Samples
std.slop.contradiction	0%	3
std.slop.format	100%	3
std.slop.hallucination	0%	3
std.slop.relevance	0%	3
std.slop.topic_drift	0%	3
std.slop.uncertainty	0%	3

Compile gates

G0 — status=complete
G1 — all required standards present
G2 — json break=0%/3
G3 — composite=100%
G4 — safety fn rate=0%
G5 — safety fp rate=0%
G6 — latency_p95 ms=9363% cap ms=30000% size band=9%to70b
G6b — throughput_p50 tps=13.5% floor tps=15% size band=9%to70b
G7 — cost proxy=0% median comparable=0%
G8 — provider error rate=0%
G9 — no easy_only_cap

Safety gates

Not tested

Weakness tags

weakness:math_halluc (error_mode / ERR_HALLUCINATION)

Not required — the safety pack could not reach this model on the provider API. This is not a behavioral safety score.

Model key: together/mistralai/mistral-small-24b-instruct-2501@created-1738246136

Model ID: mistralai/mistral-small-24b-instruct-2501

Size band: 9to70b

Provenance: provider_standard

Capability pack: bench-pack-v2

Safety pack: bench-pack-safety-v2

Latency P95: 127.4 ms

Throughput P50: — tps

Cost / task: $0

Standards

Standard	Composite	Tier	Easy	Medium	Hard	Easy-only cap
std.coding	0%	worst	0	0	0	No
std.instruction_follow	0%	worst	—	—	—	No
std.json_structured	0%	worst	0	0	0	No
std.long_context	0%	worst	—	—	—	No
std.math	0%	worst	0	0	0	No
std.multiturn	0%	worst	—	—	—	No
std.reasoning	0%	worst	0	0	0	No
std.safety_policy	0%	worst	—	—	—	No
std.slop.contradiction	0%	worst	—	—	—	No
std.slop.format	0%	worst	—	—	—	No
std.slop.hallucination	0%	worst	—	—	—	No
std.slop.relevance	0%	worst	—	—	—	No
std.slop.topic_drift	0%	worst	—	—	—	No
std.slop.uncertainty	0%	worst	—	—	—	No
std.stability.refusal	0%	worst	—	—	—	No
std.stability.repeat	0%	worst	—	—	—	No
std.stability.schema	0%	worst	—	—	—	No
std.summarization	0%	worst	0	0	0	No
std.tool_use	0%	worst	—	—	—	No
std.translation	0%	worst	—	—	—	No

Slop profiles

Standard	Violation %	Samples
std.slop.contradiction	100%	3
std.slop.format	100%	3
std.slop.hallucination	100%	3
std.slop.relevance	100%	3
std.slop.topic_drift	100%	3
std.slop.uncertainty	100%	3

Compile gates

G0 — status=complete
G1 — all required standards present
G2 — json break=0%/3
G3 — composite=0%
G4 — safety fn rate=100%
G5 — safety fp rate=100%
G6 — latency_p95 ms=127.4% cap ms=30000% size band=9%to70b
G6b — throughput_p50_tps=-1 floor tps=15% size band=9%to70b
G7 — cost proxy=0% median comparable=0%
G8 — provider error rate=100%
G9 — no easy_only_cap

Safety gates

P0 — status=complete pack=bench-pack-safety-v2
P1 — all safety standards present
P9 — scorable attempts=0%
P6 — catalog_provenance=provider_standard
P2 — skipped_no_scorable
P3 — skipped_no_scorable
P4 — skipped_no_scorable
P5 — skipped_no_scorable
P7 — skipped_no_scorable
P8 — bias_report_only=true

Weakness tags

weakness:over_refuse (safety_fp / )
weakness:under_refuse (safety_fn / )

Model key: Qwen/Qwen2.5-7B-Instruct-Turbo

Model ID: Qwen/Qwen2.5-7B-Instruct-Turbo

Size band: le8b

Provenance: provider_standard

Capability pack: bench-pack-v2

Safety pack: —

Latency P95: 5055 ms

Throughput P50: 30.4 tps

Cost / task: $0.000038

Strengths

std.coding (best)
std.instruction_follow (best)
std.json_structured (best)
std.multiturn (best)
std.safety_policy (good)
std.slop.contradiction (best)
std.slop.hallucination (best)
std.slop.relevance (best)
std.slop.topic_drift (best)
std.slop.uncertainty (best)
std.stability.refusal (best)
std.stability.repeat (best)
std.stability.schema (best)
std.summarization (good)
std.tool_use (best)

Standards

Standard	Composite	Tier	Easy	Medium	Hard	Easy-only cap
std.coding	85%	best	0	1	1	No
std.instruction_follow	100%	best	—	—	—	No
std.json_structured	100%	best	1	1	1	No
std.long_context	0%	worst	—	—	—	No
std.math	25%	worst	0	0	1	No
std.multiturn	100%	best	—	—	—	No
std.reasoning	25%	worst	0	0	1	No
std.safety_policy	83.3%	good	—	—	—	No
std.slop.contradiction	100%	best	—	—	—	No
std.slop.format	0%	worst	—	—	—	No
std.slop.hallucination	100%	best	—	—	—	No
std.slop.relevance	100%	best	—	—	—	No
std.slop.topic_drift	100%	best	—	—	—	No
std.slop.uncertainty	100%	best	—	—	—	No
std.stability.refusal	100%	best	—	—	—	No
std.stability.repeat	100%	best	—	—	—	No
std.stability.schema	100%	best	—	—	—	No
std.summarization	76.7%	good	0	1	0.67	No
std.tool_use	100%	best	—	—	—	No
std.translation	0%	worst	—	—	—	No

Slop profiles

Standard	Violation %	Samples
std.slop.contradiction	0%	3
std.slop.format	100%	3
std.slop.hallucination	0%	3
std.slop.relevance	0%	3
std.slop.topic_drift	0%	3
std.slop.uncertainty	0%	3

Compile gates

G0 — status=complete
G1 — all required standards present
G2 — json break=0%/3
G3 — composite=100%
G4 — safety fn rate=0%
G5 — safety fp rate=0%
G6 — latency_p95 ms=5055% cap ms=10000% size_band=le8b
G6b — throughput_p50 tps=30.4% floor tps=20% size_band=le8b
G7 — cost proxy=0% median comparable=0%
G8 — provider error rate=0%
G9 — no easy_only_cap

Safety gates

Not tested

Weakness tags

weakness:math_halluc (error_mode / ERR_HALLUCINATION)