{ "schema": "ai4science.rankings-model-sources.v1", "checked_date": "2026-10-08", "formula": "0.60 * GPQA percentage + 0.40 * FrontierMath percentage", "fixed_reference_usage": { "uncached_input_tokens": 750000, "output_tokens_including_reasoning": 250000 }, "records": [ { "model": { "model_version": "gpt-6-astra_max", "scientific_reasoning_beta": 94.93580542264753, "display_score": 94.9, "gpqa_percent": 95.77020202020202, "frontiermath_percent": 93.6842105263158, "core_coverage": "2/2", "broader_science_scope": "scientific QA + research math only; code/workflow not scored", "se_upper_bound": 1.39917025763497, "approx_95_low": 92.193431717683, "approx_95_high": 97.67817912761207, "run_started_at_min": "2026-08-30T14:57:43.000Z", "run_started_at_max": "2026-08-30T14:57:43.000Z", "run_time_relation": "pre_release", "identity_tier": "I2", "evaluator_count": 1, "score_status": "provisional_two_benchmark_composite", "official_provider_source": "https://developers.openai.com/api/docs/models", "point_order": 1, "conservative_rank_min": 1, "conservative_rank_max": 5, "model_key": "gpt-6-astra_max", "display_name": "GPT-6 Astra", "provider": "OpenAI", "official_api_model_id": "gpt-6-astra", "effort": "max", "effort_verified": true, "effort_is_separate_from_tariff": true, "currency": "USD", "input_usd_per_million": 10, "output_usd_per_million": 50, "reference_cost_usd": 20.0, "reference_cost_display": "$20.00", "reference_cost_relative_to_sol": 5.0, "scope": "OpenAI direct API, Standard processing, ordinary context, no regional-processing premium", "status": "verified_current_official_api_tariff", "checked_date": "2026-10-08", "checked_at": "2026-10-08T03:05:56Z", "source": "https://developers.openai.com/api/docs/models/gpt-6-astra", "sources": [ { "url": "https://developers.openai.com/api/docs/models/gpt-6-astra", "role": "tariff" }, { "url": "https://developers.openai.com/api/docs/models/gpt-6-astra", "role": "official_identity" }, { "url": "https://developers.openai.com/api/docs/models/gpt-6-astra", "role": "supported_effort" }, { "url": "https://developers.openai.com/api/docs/guides/reasoning", "role": "reasoning_billing" } ], "conditions": { "processing": "Standard", "ordinary_context_max_input_tokens": 272000, "long_context": { "trigger": "input_tokens > 272000", "all_request_tokens_repriced": true, "input_usd_per_million": 20, "output_usd_per_million": 75.0 }, "reasoning_tokens_billed_as": "output", "regional_processing": { "multiplier": 1.1, "excluded_from_reference": true, "source": "https://developers.openai.com/api/docs/pricing" }, "batch_flex_fast_ultrafast_excluded": true, "caching_excluded": true, "tools_and_compute_fees_excluded": true }, "limitations": [ "Reference token tariff only; not measured evaluation spend, cost per answer, or cost per successful task.", "Published API identity and supported effort are verified, but the benchmark source does not independently verify an immutable provider snapshot or per-row budgets (I2).", "Fixed billable-token counts do not represent equal text, reasoning work, latency, retries, or task quality across tokenizers.", "The Astra benchmark run is pre-release; current commercial rates cannot establish its evaluated preview snapshot or historical charge." ], "benchmark_identity_status": "source_label_matches_official_model; immutable_evaluated_snapshot_unverified", "benchmark_identity_tier": "I2", "price_is_historical_benchmark_price": false, "on_point_estimate_cost_capability_frontier": true, "name": "GPT-6 Astra", "hint": "高分候选 · 预发布证据", "n": 0, "cache_pricing": { "model_key": "gpt-6-astra_max", "official_api_model_id": "gpt-6-astra", "display_name": "GPT-6 Astra", "scope": "OpenAI direct API; Standard; ordinary context; no regional premium.", "input_usd_per_million": 10, "output_usd_per_million": 50, "cached_read_usd_per_million": 1, "cache_write_usd_per_million": 12.5, "cache_write": { "usd_per_million": 12.5, "input_multiplier": 1.25, "ttl": "30m" }, "cache_storage": null, "conditions": { "ordinary_context_max_input_tokens": 272000, "long_context": { "trigger": "input_tokens > 272000", "all_request_tokens_repriced": true, "input_usd_per_million": 20, "output_usd_per_million": 75, "cached_read_usd_per_million": 2, "cache_write_usd_per_million": 25 }, "cache_minimum_visible_input_tokens": 1024, "ttl_refreshes_on_reuse": true, "minimum_lifetime_after_write_or_reuse_minutes": 30, "promotional_period": null }, "status": "verified_current_official_cached_read_tariff", "sourceURL": "https://developers.openai.com/api/docs/models/gpt-6-astra", "sources": [ "https://developers.openai.com/api/docs/models/gpt-6-astra", "https://developers.openai.com/api/docs/guides/prompt-caching" ], "notes": [ "Cache-read and write prices are explicitly listed by the model page.", "Writes and reads are distinct billing categories. Ordinary uncached reference-cost chart is unchanged." ], "checkedAt": "2026-10-08T03:22:13Z" } }, "benchmark_records": [ { "benchmark": "GPQA Diamond", "mean_score": 0.95770202, "mean_score_display": "0.95770202", "standard_error_percentage_points": 1.37, "se_display": "1.370", "started_at": "2026-08-30T14:57:43.000Z", "record_id": "3h8STJpG2VzY9LiBUzwotv", "csv": "gpqa_diamond.csv", "protocol_url": "https://epoch.ai/benchmarks/gpqa-diamond" }, { "benchmark": "FrontierMath v2", "mean_score": 0.93684211, "mean_score_display": "0.93684211", "standard_error_percentage_points": 1.443, "se_display": "1.443", "started_at": "2026-08-30T14:57:43.000Z", "record_id": "csK2fe5T8p2R7WgoJUu6bZ", "csv": "frontiermath_tiers_1_3_v2.csv", "protocol_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2" } ] }, { "model": { "model_version": "gpt-6.1-sol_max", "scientific_reasoning_beta": 94.7085326953748, "display_score": 94.7, "gpqa_percent": 95.39141414141415, "frontiermath_percent": 93.6842105263158, "core_coverage": "2/2", "broader_science_scope": "scientific QA + research math only; code/workflow not scored", "se_upper_bound": 1.404206438835633, "approx_95_low": 91.95628807525696, "approx_95_high": 97.46077731549263, "run_started_at_min": "2026-09-29T22:00:34.000Z", "run_started_at_max": "2026-09-29T22:00:34.000Z", "run_time_relation": "at_or_after_release", "identity_tier": "I2", "evaluator_count": 1, "score_status": "provisional_two_benchmark_composite", "official_provider_source": "https://developers.openai.com/api/docs/models", "point_order": 2, "conservative_rank_min": 1, "conservative_rank_max": 5, "model_key": "gpt-6.1-sol_max", "display_name": "GPT-6.1 Sol", "provider": "OpenAI", "official_api_model_id": "gpt-6.1-sol", "effort": "max", "effort_verified": true, "effort_is_separate_from_tariff": true, "currency": "USD", "input_usd_per_million": 2, "output_usd_per_million": 10, "reference_cost_usd": 4.0, "reference_cost_display": "$4.00", "reference_cost_relative_to_sol": 1.0, "scope": "OpenAI direct API, Standard processing, ordinary context, no regional-processing premium", "status": "verified_current_official_api_tariff", "checked_date": "2026-10-08", "checked_at": "2026-10-08T03:05:56Z", "source": "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "sources": [ { "url": "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "role": "tariff" }, { "url": "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "role": "official_identity" }, { "url": "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "role": "supported_effort" }, { "url": "https://developers.openai.com/api/docs/guides/reasoning", "role": "reasoning_billing" } ], "conditions": { "processing": "Standard", "ordinary_context_max_input_tokens": 272000, "long_context": { "trigger": "input_tokens > 272000", "all_request_tokens_repriced": true, "input_usd_per_million": 4, "output_usd_per_million": 15.0 }, "reasoning_tokens_billed_as": "output", "regional_processing": { "multiplier": 1.1, "excluded_from_reference": true, "source": "https://developers.openai.com/api/docs/pricing" }, "batch_flex_fast_ultrafast_excluded": true, "caching_excluded": true, "tools_and_compute_fees_excluded": true }, "limitations": [ "Reference token tariff only; not measured evaluation spend, cost per answer, or cost per successful task.", "Published API identity and supported effort are verified, but the benchmark source does not independently verify an immutable provider snapshot or per-row budgets (I2).", "Fixed billable-token counts do not represent equal text, reasoning work, latency, retries, or task quality across tokenizers." ], "benchmark_identity_status": "source_label_matches_official_model; immutable_evaluated_snapshot_unverified", "benchmark_identity_tier": "I2", "price_is_historical_benchmark_price": false, "on_point_estimate_cost_capability_frontier": true, "name": "GPT-6.1 Sol", "hint": "高分候选 · Astra 的 1/5 成本", "n": 1, "cache_pricing": { "model_key": "gpt-6.1-sol_max", "official_api_model_id": "gpt-6.1-sol", "display_name": "GPT-6.1 Sol", "scope": "OpenAI direct API; Standard; ordinary context; no regional premium.", "input_usd_per_million": 2, "output_usd_per_million": 10, "cached_read_usd_per_million": 0.1, "cache_write_usd_per_million": 2.5, "cache_write": { "usd_per_million": 2.5, "input_multiplier": 1.25, "ttl": "30m" }, "cache_storage": null, "conditions": { "ordinary_context_max_input_tokens": 272000, "long_context": { "trigger": "input_tokens > 272000", "all_request_tokens_repriced": true, "input_usd_per_million": 4, "output_usd_per_million": 15, "cached_read_usd_per_million": 0.2, "cache_write_usd_per_million": 5 }, "cache_minimum_visible_input_tokens": 1024, "ttl_refreshes_on_reuse": true, "minimum_lifetime_after_write_or_reuse_minutes": 30, "promotional_period": null }, "status": "verified_current_official_cached_read_tariff", "sourceURL": "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "sources": [ "https://developers.openai.com/api/docs/models/gpt-6.1-sol", "https://developers.openai.com/api/docs/guides/prompt-caching" ], "notes": [ "Official model page lists $0.10 cached input, equal to 5% of uncached input; do not use a generic 10% assumption." ], "checkedAt": "2026-10-08T03:22:13Z" } }, "benchmark_records": [ { "benchmark": "GPQA Diamond", "mean_score": 0.95391414, "mean_score_display": "0.95391414", "standard_error_percentage_points": 1.378, "se_display": "1.378", "started_at": "2026-09-29T22:00:34.000Z", "record_id": "LspviLnbeZ82WCngWp6rwR", "csv": "gpqa_diamond.csv", "protocol_url": "https://epoch.ai/benchmarks/gpqa-diamond" }, { "benchmark": "FrontierMath v2", "mean_score": 0.93684211, "mean_score_display": "0.93684211", "standard_error_percentage_points": 1.443, "se_display": "1.443", "started_at": "2026-09-29T22:00:34.000Z", "record_id": "aBoQLaqnXsLPrsbN8wYgUp", "csv": "frontiermath_tiers_1_3_v2.csv", "protocol_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2" } ] }, { "model": { "model_version": "claude-sonnet-5-5_max", "scientific_reasoning_beta": 92.85725677830942, "display_score": 92.9, "gpqa_percent": 95.58080808080808, "frontiermath_percent": 88.7719298245614, "core_coverage": "2/2", "broader_science_scope": "scientific QA + research math only; code/workflow not scored", "se_upper_bound": 1.571980457292242, "approx_95_low": 89.77617508201662, "approx_95_high": 95.93833847460222, "run_started_at_min": "2026-09-29T15:15:14.000Z", "run_started_at_max": "2026-09-29T15:15:14.000Z", "run_time_relation": "at_or_after_release", "identity_tier": "I2", "evaluator_count": 1, "score_status": "provisional_two_benchmark_composite", "official_provider_source": "https://www.anthropic.com/claude-sonnet-5-5", "point_order": 3, "conservative_rank_min": 1, "conservative_rank_max": 6, "model_key": "claude-sonnet-5-5_max", "display_name": "Claude Sonnet 5.5", "provider": "Anthropic", "official_api_model_id": "claude-sonnet-5-5", "effort": "max", "effort_verified": true, "effort_is_separate_from_tariff": true, "currency": "USD", "input_usd_per_million": 2, "output_usd_per_million": 10, "reference_cost_usd": 4.0, "reference_cost_display": "$4.00", "reference_cost_relative_to_sol": 1.0, "scope": "Anthropic direct Claude API, standard speed, default global inference", "status": "verified_current_official_api_tariff", "checked_date": "2026-10-08", "checked_at": "2026-10-08T03:05:56Z", "source": "https://platform.claude.com/docs/en/about-claude/pricing", "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "role": "tariff" }, { "url": "https://platform.claude.com/docs/en/models/sonnet-5-5/overview", "role": "official_identity" }, { "url": "https://platform.claude.com/docs/en/build-with-claude/effort", "role": "supported_effort" }, { "url": "https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost", "role": "reasoning_billing" } ], "conditions": { "processing": "standard speed", "inference_geo": "global", "long_context": { "premium": false, "context_window_tokens": 1000000, "explanation": "The full supported context window uses standard per-token pricing." }, "reasoning_tokens_billed_as": "output", "us_only_inference": { "multiplier": 1.1, "excluded_from_reference": true }, "batch_fast_mode_excluded": true, "caching_excluded": true, "tools_and_compute_fees_excluded": true }, "limitations": [ "Reference token tariff only; not measured evaluation spend, cost per answer, or cost per successful task.", "Published API identity and supported effort are verified, but the benchmark source does not independently verify an immutable provider snapshot or per-row budgets (I2).", "Fixed billable-token counts do not represent equal text, reasoning work, latency, retries, or task quality across tokenizers.", "Partner-operated Bedrock and Google Cloud have their own pricing; the selected route is the first-party API." ], "benchmark_identity_status": "source_label_matches_official_model; immutable_evaluated_snapshot_unverified", "benchmark_identity_tier": "I2", "price_is_historical_benchmark_price": false, "on_point_estimate_cost_capability_frontier": false, "name": "Claude Sonnet 5.5", "hint": "同价参考：Sol 点分更高", "n": 2, "cache_pricing": { "model_key": "claude-sonnet-5-5_max", "official_api_model_id": "claude-sonnet-5-5", "display_name": "Claude Sonnet 5.5", "scope": "Anthropic direct Claude API; standard speed; global routing.", "input_usd_per_million": 2, "output_usd_per_million": 10, "cached_read_usd_per_million": 0.1, "cache_write_usd_per_million": null, "cache_write_5min_usd_per_million": 2.5, "cache_write_1h_usd_per_million": 4, "cache_write": { "five_minute": { "usd_per_million": 2.5, "input_multiplier": 1.25, "ttl": "5m" }, "one_hour": { "usd_per_million": 4, "input_multiplier": 2, "ttl": "1h" } }, "cache_storage": null, "conditions": { "full_context_window_tokens": 1000000, "long_context_premium": false, "cache_read_input_multiplier": 0.05, "us_only_inference_multiplier": 1.1, "selected_inference_geo": "global", "promotional_period": null }, "status": "verified_current_official_cached_read_tariff", "sourceURL": "https://platform.claude.com/docs/en/about-claude/pricing", "sources": [ "https://platform.claude.com/docs/en/about-claude/pricing" ], "notes": [ "Hits and refreshes cost $0.10; distinct 5-minute and 1-hour writes cost $2.50 and $4.", "Standard caching prices apply throughout the 1M-token context window." ], "checkedAt": "2026-10-08T03:22:13Z" } }, "benchmark_records": [ { "benchmark": "GPQA Diamond", "mean_score": 0.95580808, "mean_score_display": "0.95580808", "standard_error_percentage_points": 1.371, "se_display": "1.371", "started_at": "2026-09-29T15:15:14.000Z", "record_id": "QPk8jbJvo4986sbWtdte6J", "csv": "gpqa_diamond.csv", "protocol_url": "https://epoch.ai/benchmarks/gpqa-diamond" }, { "benchmark": "FrontierMath v2", "mean_score": 0.8877193, "mean_score_display": "0.88771930", "standard_error_percentage_points": 1.873, "se_display": "1.873", "started_at": "2026-09-29T15:15:14.000Z", "record_id": "4tZWwV9tqEgsCaKeSuSoG4", "csv": "frontiermath_tiers_1_3_v2.csv", "protocol_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2" } ] }, { "model": { "model_version": "claude-opus-5-5_max", "scientific_reasoning_beta": 90.84728867623605, "display_score": 90.8, "gpqa_percent": 90.59343434343434, "frontiermath_percent": 91.22807017543859, "core_coverage": "2/2", "broader_science_scope": "scientific QA + research math only; code/workflow not scored", "se_upper_bound": 1.837512225749212, "approx_95_low": 87.24576471376758, "approx_95_high": 94.44881263870451, "run_started_at_min": "2026-09-22T18:27:39.000Z", "run_started_at_max": "2026-09-22T18:27:39.000Z", "run_time_relation": "at_or_after_release", "identity_tier": "I2", "evaluator_count": 1, "score_status": "provisional_two_benchmark_composite", "official_provider_source": "https://www.anthropic.com/claude-opus-5-5", "point_order": 4, "conservative_rank_min": 1, "conservative_rank_max": 6, "model_key": "claude-opus-5-5_max", "display_name": "Claude Opus 5.5", "provider": "Anthropic", "official_api_model_id": "claude-opus-5-5", "effort": "max", "effort_verified": true, "effort_is_separate_from_tariff": true, "currency": "USD", "input_usd_per_million": 4, "output_usd_per_million": 20, "reference_cost_usd": 8.0, "reference_cost_display": "$8.00", "reference_cost_relative_to_sol": 2.0, "scope": "Anthropic direct Claude API, standard speed, default global inference", "status": "verified_current_official_api_tariff", "checked_date": "2026-10-08", "checked_at": "2026-10-08T03:05:56Z", "source": "https://platform.claude.com/docs/en/about-claude/pricing", "sources": [ { "url": "https://platform.claude.com/docs/en/about-claude/pricing", "role": "tariff" }, { "url": "https://platform.claude.com/docs/en/models/opus-5-5/overview", "role": "official_identity" }, { "url": "https://platform.claude.com/docs/en/build-with-claude/effort", "role": "supported_effort" }, { "url": "https://platform.claude.com/docs/en/build-with-claude/thinking-steering-and-cost", "role": "reasoning_billing" } ], "conditions": { "processing": "standard speed", "inference_geo": "global", "long_context": { "premium": false, "context_window_tokens": 1000000, "explanation": "The full supported context window uses standard per-token pricing." }, "reasoning_tokens_billed_as": "output", "us_only_inference": { "multiplier": 1.1, "excluded_from_reference": true }, "batch_fast_mode_excluded": true, "caching_excluded": true, "tools_and_compute_fees_excluded": true }, "limitations": [ "Reference token tariff only; not measured evaluation spend, cost per answer, or cost per successful task.", "Published API identity and supported effort are verified, but the benchmark source does not independently verify an immutable provider snapshot or per-row budgets (I2).", "Fixed billable-token counts do not represent equal text, reasoning work, latency, retries, or task quality across tokenizers.", "Partner-operated Bedrock and Google Cloud have their own pricing; the selected route is the first-party API." ], "benchmark_identity_status": "source_label_matches_official_model; immutable_evaluated_snapshot_unverified", "benchmark_identity_tier": "I2", "price_is_historical_benchmark_price": false, "on_point_estimate_cost_capability_frontier": false, "name": "Claude Opus 5.5", "hint": "低价参考：Sol 点分更高", "n": 3, "cache_pricing": { "model_key": "claude-opus-5-5_max", "official_api_model_id": "claude-opus-5-5", "display_name": "Claude Opus 5.5", "scope": "Anthropic direct Claude API; standard speed; global routing.", "input_usd_per_million": 4, "output_usd_per_million": 20, "cached_read_usd_per_million": 0.2, "cache_write_usd_per_million": null, "cache_write_5min_usd_per_million": 5, "cache_write_1h_usd_per_million": 8, "cache_write": { "five_minute": { "usd_per_million": 5, "input_multiplier": 1.25, "ttl": "5m" }, "one_hour": { "usd_per_million": 8, "input_multiplier": 2, "ttl": "1h" } }, "cache_storage": null, "conditions": { "full_context_window_tokens": 1000000, "long_context_premium": false, "cache_read_input_multiplier": 0.05, "us_only_inference_multiplier": 1.1, "selected_inference_geo": "global", "promotional_period": null }, "status": "verified_current_official_cached_read_tariff", "sourceURL": "https://platform.claude.com/docs/en/about-claude/pricing", "sources": [ "https://platform.claude.com/docs/en/about-claude/pricing" ], "notes": [ "Hits and refreshes cost $0.20; distinct 5-minute and 1-hour writes cost $5 and $8.", "Fast mode and regional premiums are excluded." ], "checkedAt": "2026-10-08T03:22:13Z" } }, "benchmark_records": [ { "benchmark": "GPQA Diamond", "mean_score": 0.90593434, "mean_score_display": "0.90593434", "standard_error_percentage_points": 1.943, "se_display": "1.943", "started_at": "2026-09-22T18:27:39.000Z", "record_id": "6gmbahPSxHf9xVhAYqpnhi", "csv": "gpqa_diamond.csv", "protocol_url": "https://epoch.ai/benchmarks/gpqa-diamond" }, { "benchmark": "FrontierMath v2", "mean_score": 0.9122807, "mean_score_display": "0.91228070", "standard_error_percentage_points": 1.679, "se_display": "1.679", "started_at": "2026-09-22T18:27:39.000Z", "record_id": "LPxhvBNEn7S3xkRZUgcvc9", "csv": "frontiermath_tiers_1_3_v2.csv", "protocol_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2" } ] }, { "model": { "model_version": "gemini-3.8-flash_high", "scientific_reasoning_beta": 84.60326953748006, "display_score": 84.6, "gpqa_percent": 95.39141414141415, "frontiermath_percent": 68.42105263157895, "core_coverage": "2/2", "broader_science_scope": "scientific QA + research math only; code/workflow not scored", "se_upper_bound": 1.9439840862645115, "approx_95_low": 80.79306072840161, "approx_95_high": 88.4134783465585, "run_started_at_min": "2026-09-02T17:15:36.000Z", "run_started_at_max": "2026-09-02T17:15:36.000Z", "run_time_relation": "at_or_after_release", "identity_tier": "I2", "evaluator_count": 1, "score_status": "provisional_two_benchmark_composite", "official_provider_source": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash", "point_order": 5, "conservative_rank_min": 1, "conservative_rank_max": 7, "model_key": "gemini-3.8-flash_high", "display_name": "Gemini 3.8 Flash", "provider": "Google DeepMind", "official_api_model_id": "gemini-3.8-flash", "effort": "high", "effort_verified": true, "effort_is_separate_from_tariff": true, "currency": "USD", "input_usd_per_million": 0.75, "output_usd_per_million": 3.75, "reference_cost_usd": 1.5, "reference_cost_display": "$1.50", "reference_cost_relative_to_sol": 0.375, "scope": "Google Gemini Developer API paid tier, Standard processing, current introductory rate", "status": "verified_current_official_introductory_api_tariff", "checked_date": "2026-10-08", "checked_at": "2026-10-08T03:05:56Z", "source": "https://ai.google.dev/gemini-api/docs/pricing", "sources": [ { "url": "https://ai.google.dev/gemini-api/docs/pricing", "role": "tariff" }, { "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash", "role": "official_identity" }, { "url": "https://ai.google.dev/gemini-api/docs/thinking", "role": "supported_effort" }, { "url": "https://ai.google.dev/gemini-api/docs/pricing", "role": "reasoning_billing" } ], "conditions": { "processing": "Standard", "paid_tier": true, "current_price_type": "introductory", "current_price_valid_through": "2026-12-31", "scheduled_future_tariff": { "effective_date": "2027-01-01", "input_usd_per_million": 1.5, "output_usd_per_million": 7.5, "reference_cost_usd": 3 }, "long_context": { "premium_not_listed_for_selected_model_on_developer_api_price_table": true }, "reasoning_tokens_billed_as": "output", "batch_flex_priority_free_tier_excluded": true, "caching_excluded": true, "grounding_tools_and_storage_fees_excluded": true, "cloud_platform_alternative": { "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", "non_global_multiplier": 1.1, "excluded_from_reference": true } }, "limitations": [ "Reference token tariff only; not measured evaluation spend, cost per answer, or cost per successful task.", "Published API identity and supported effort are verified, but the benchmark source does not independently verify an immutable provider snapshot or per-row budgets (I2).", "Fixed billable-token counts do not represent equal text, reasoning work, latency, retries, or task quality across tokenizers.", "The current introductory price is time-limited; do not present it as a permanent standing tariff." ], "benchmark_identity_status": "source_label_matches_official_model; immutable_evaluated_snapshot_unverified", "benchmark_identity_tier": "I2", "price_is_historical_benchmark_price": false, "on_point_estimate_cost_capability_frontier": true, "name": "Gemini 3.8 Flash", "hint": "低价候选 · 当前优惠价", "n": 4, "cache_pricing": { "model_key": "gemini-3.8-flash_high", "official_api_model_id": "gemini-3.8-flash", "display_name": "Gemini 3.8 Flash", "scope": "Gemini Developer API; paid tier; Standard; current introductory pricing.", "input_usd_per_million": 0.75, "output_usd_per_million": 3.75, "cached_read_usd_per_million": 0.075, "cache_write_usd_per_million": null, "cache_write": null, "cache_storage_usd_per_million_tokens_per_hour": 0.5, "cache_storage": { "usd_per_million_tokens_per_hour": 0.5, "applies_to": "Explicit cache storage; separate from cached-read token tariff.", "valid_through": "2026-12-31" }, "conditions": { "current_price_type": "introductory", "current_price_valid_through": "2026-12-31", "scheduled_future_tariff": { "effective_date": "2027-01-01", "input_usd_per_million": 1.5, "output_usd_per_million": 7.5, "cached_read_usd_per_million": 0.15, "cache_storage_usd_per_million_tokens_per_hour": 1 }, "implicit_cache_min_input_tokens": 4096, "implicit_cache_enabled_by_default": true, "explicit_cache_api": "generateContent", "interactions_api_explicit_cache_supported": false, "long_context_premium_not_listed": true }, "status": "verified_current_official_introductory_cached_read_tariff", "sourceURL": "https://ai.google.dev/gemini-api/docs/pricing", "sources": [ "https://ai.google.dev/gemini-api/docs/pricing", "https://ai.google.dev/gemini-api/docs/caching" ], "notes": [ "Standard cached-read tariff is $0.075 through 2026-12-31; explicit storage costs $0.50 per million tokens per hour.", "No separate model cache-write tariff is listed in the checked table; null is not a zero-price claim." ], "checkedAt": "2026-10-08T03:22:13Z" } }, "benchmark_records": [ { "benchmark": "GPQA Diamond", "mean_score": 0.95391414, "mean_score_display": "0.95391414", "standard_error_percentage_points": 1.401, "se_display": "1.401", "started_at": "2026-09-02T17:15:36.000Z", "record_id": "Br3mjs2GrYcCwX9AMwcihU", "csv": "gpqa_diamond.csv", "protocol_url": "https://epoch.ai/benchmarks/gpqa-diamond" }, { "benchmark": "FrontierMath v2", "mean_score": 0.68421053, "mean_score_display": "0.68421053", "standard_error_percentage_points": 2.758, "se_display": "2.758", "started_at": "2026-09-02T17:15:36.000Z", "record_id": "Utgs6DuaQDDdi98ZWH22qL", "csv": "frontiermath_tiers_1_3_v2.csv", "protocol_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2" } ] }, { "model": { "model_version": "qwen3.8-max-0902_xhigh", "scientific_reasoning_beta": 81.6244019138756, "display_score": 81.6, "gpqa_percent": 92.2979797979798, "frontiermath_percent": 65.6140350877193, "core_coverage": "2/2", "broader_science_scope": "scientific QA + research math only; code/workflow not scored", "se_upper_bound": 2.1624291793592456, "approx_95_low": 77.38604072233149, "approx_95_high": 85.86276310541972, "run_started_at_min": "2026-09-02T18:04:58.000Z", "run_started_at_max": "2026-09-02T18:04:58.000Z", "run_time_relation": "at_or_after_release", "identity_tier": "I2", "evaluator_count": 1, "score_status": "provisional_two_benchmark_composite", "official_provider_source": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "point_order": 6, "conservative_rank_min": 3, "conservative_rank_max": 7, "model_key": "qwen3.8-max-0902_xhigh", "display_name": "Qwen3.8 Max 0902", "provider": "Alibaba", "official_api_model_id": "qwen3.8-max-0902", "effort": "xhigh", "effort_verified": true, "effort_is_separate_from_tariff": true, "currency": "USD", "input_usd_per_million": 2, "output_usd_per_million": 6, "reference_cost_usd": 3.0, "reference_cost_display": "$3.00", "reference_cost_relative_to_sol": 0.75, "scope": "Alibaba Cloud Model Studio, Singapore region, International deployment, ordinary pay-as-you-go API", "status": "verified_current_official_api_tariff", "checked_date": "2026-10-08", "checked_at": "2026-10-08T03:05:56Z", "source": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "sources": [ { "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "role": "tariff" }, { "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "role": "official_identity" }, { "url": "https://www.alibabacloud.com/help/en/model-studio/qwen-api-via-openai-chat-completions", "role": "supported_effort" }, { "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "role": "reasoning_billing" } ], "conditions": { "processing": "ordinary pay-as-you-go", "selected_region": "Singapore", "selected_deployment_scope": "International", "request_input_token_tariff_range": { "lower_exclusive": 0, "upper_inclusive": 1000000 }, "reasoning_tokens_billed_as": "output (chain of thought + answer)", "same_tariff_for_thinking_and_non_thinking": true, "caching_excluded": true, "batch_discounts_and_free_quota_excluded": true, "tools_and_compute_fees_excluded": true, "regional_alternatives": [ { "region": "Hong Kong / Frankfurt / Virginia / Tokyo", "deployment_scope": "Global", "input_usd_per_million": 1.65, "output_usd_per_million": 4.951, "reference_cost_usd": 2.47525, "source": "https://www.alibabacloud.com/help/en/model-studio/model-pricing" } ], "official_price_page_updated": "2026-10-06" }, "limitations": [ "Reference token tariff only; not measured evaluation spend, cost per answer, or cost per successful task.", "Published API identity and supported effort are verified, but the benchmark source does not independently verify an immutable provider snapshot or per-row budgets (I2).", "Fixed billable-token counts do not represent equal text, reasoning work, latency, retries, or task quality across tokenizers.", "The selected International route is a disclosed reference choice; Global routes have different official USD list prices.", "No Qwen preview alias is substituted for the exact 0902 ID." ], "benchmark_identity_status": "source_label_matches_official_model; immutable_evaluated_snapshot_unverified", "benchmark_identity_tier": "I2", "price_is_historical_benchmark_price": false, "on_point_estimate_cost_capability_frontier": false, "name": "Qwen3.8 Max 0902", "hint": "低价参考：Flash 点分更高", "n": 5, "cache_pricing": { "model_key": "qwen3.8-max-0902_xhigh", "official_api_model_id": "qwen3.8-max-0902", "display_name": "Qwen3.8 Max 0902", "scope": "Alibaba Cloud Model Studio; Singapore; International deployment; ordinary pay-as-you-go.", "input_usd_per_million": 2, "output_usd_per_million": 6, "cached_read_usd_per_million": null, "cached_read_display_zh": "未公开", "cache_write_usd_per_million": 2.5, "cache_write": { "explicit": { "usd_per_million": 2.5, "input_multiplier": 1.25, "ttl": "5m" }, "implicit": { "usd_per_million": 2, "input_multiplier": 1 } }, "cache_storage": null, "conditions": { "selected_region": "Singapore", "selected_deployment_scope": "International", "request_input_token_tariff_range": { "lower_exclusive": 0, "upper_inclusive": 1000000 }, "explicit_cache_supported": true, "implicit_cache_supported": true, "explicit_cached_read_usd_per_million": null, "implicit_cached_read_usd_per_million": null, "cache_minimum_input_tokens": 1024, "explicit_cache_ttl": "5m", "explicit_cache_ttl_refreshes_on_hit": true, "promotional_period": null }, "status": "exact_cached_read_rate_not_publicly_listed_console_required", "sourceURL": "https://www.alibabacloud.com/help/en/model-studio/context-cache", "sources": [ "https://www.alibabacloud.com/help/en/model-studio/context-cache", "https://www.alibabacloud.com/help/en/model-studio/model-pricing" ], "notes": [ "Official cache docs explicitly exclude qwen3.8-max-0902 from generic 10% explicit and 20% implicit read rates; exact rates require the Model Studio console.", "No account access was used. Explicit creation is 125% of the same International input tariff; implicit creation uses ordinary input pricing." ], "checkedAt": "2026-10-08T03:22:13Z" } }, "benchmark_records": [ { "benchmark": "GPQA Diamond", "mean_score": 0.9229798, "mean_score_display": "0.92297980", "standard_error_percentage_points": 1.725, "se_display": "1.725", "started_at": "2026-09-02T18:04:58.000Z", "record_id": "VzeQXKBH9maR2nDcZJ8b7S", "csv": "gpqa_diamond.csv", "protocol_url": "https://epoch.ai/benchmarks/gpqa-diamond" }, { "benchmark": "FrontierMath v2", "mean_score": 0.65614035, "mean_score_display": "0.65614035", "standard_error_percentage_points": 2.819, "se_display": "2.819", "started_at": "2026-09-02T18:04:58.000Z", "record_id": "FpQf5xBwJPb4ZEo6SgdKWq", "csv": "frontiermath_tiers_1_3_v2.csv", "protocol_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2" } ] }, { "model": { "model_version": "grok-4.7_xhigh", "scientific_reasoning_beta": 76.83692185007975, "display_score": 76.8, "gpqa_percent": 92.73989898989899, "frontiermath_percent": 52.98245614035088, "core_coverage": "2/2", "broader_science_scope": "scientific QA + research math only; code/workflow not scored", "se_upper_bound": 2.115017939408082, "approx_95_low": 72.6914866888399, "approx_95_high": 80.98235701131959, "run_started_at_min": "2026-09-22T21:49:22.000Z", "run_started_at_max": "2026-09-22T21:49:22.000Z", "run_time_relation": "at_or_after_release", "identity_tier": "I2", "evaluator_count": 1, "score_status": "provisional_two_benchmark_composite", "official_provider_source": "https://docs.x.ai/developers/models/grok-4.7", "point_order": 7, "conservative_rank_min": 5, "conservative_rank_max": 7, "model_key": "grok-4.7_xhigh", "display_name": "Grok 4.7", "provider": "xAI", "official_api_model_id": "grok-4.7", "effort": "xhigh", "effort_verified": true, "effort_is_separate_from_tariff": true, "currency": "USD", "input_usd_per_million": 2, "output_usd_per_million": 6, "reference_cost_usd": 3.0, "reference_cost_display": "$3.00", "reference_cost_relative_to_sol": 0.75, "scope": "xAI direct public API, standard processing, global endpoint, ordinary context", "status": "verified_current_official_api_tariff", "checked_date": "2026-10-08", "checked_at": "2026-10-08T03:05:56Z", "source": "https://docs.x.ai/developers/pricing", "sources": [ { "url": "https://docs.x.ai/developers/pricing", "role": "tariff" }, { "url": "https://docs.x.ai/developers/models/grok-4.7", "role": "official_identity" }, { "url": "https://docs.x.ai/developers/models/grok-4.7", "role": "supported_effort" }, { "url": "https://docs.x.ai/developers/advanced-api-usage/prompt-caching/usage-and-pricing", "role": "reasoning_billing" } ], "conditions": { "processing": "standard", "reference_input_tokens_per_request_below": 200000, "long_context": { "trigger": "prompt_tokens >= 200000 in main pricing table", "all_request_tokens_repriced": true, "input_usd_per_million": 4, "output_usd_per_million": 12, "boundary_note": "The model-detail page says exceeds 200K; this reference stays well below the boundary." }, "reasoning_tokens_billed_as": "output / completion", "us_regional_endpoint": { "multiplier": 1.1, "excluded_from_reference": true }, "priority_fast_variant_excluded": true, "caching_excluded": true, "tools_and_compute_fees_excluded": true, "batch_supported": false }, "limitations": [ "Reference token tariff only; not measured evaluation spend, cost per answer, or cost per successful task.", "Published API identity and supported effort are verified, but the benchmark source does not independently verify an immutable provider snapshot or per-row budgets (I2).", "Fixed billable-token counts do not represent equal text, reasoning work, latency, retries, or task quality across tokenizers." ], "benchmark_identity_status": "source_label_matches_official_model; immutable_evaluated_snapshot_unverified", "benchmark_identity_tier": "I2", "price_is_historical_benchmark_price": false, "on_point_estimate_cost_capability_frontier": false, "name": "Grok 4.7", "hint": "低价参考：Flash 点分更高", "n": 6, "cache_pricing": { "model_key": "grok-4.7_xhigh", "official_api_model_id": "grok-4.7", "display_name": "Grok 4.7", "scope": "xAI direct public API; standard; global endpoint; ordinary context.", "input_usd_per_million": 2, "output_usd_per_million": 6, "cached_read_usd_per_million": 0.5, "cache_write_usd_per_million": null, "cache_write": null, "cache_storage": null, "conditions": { "ordinary_context_prompt_tokens_below": 200000, "long_context": { "trigger": "prompt_tokens >= 200000 in pricing table", "cached_tokens_included_in_threshold": true, "all_request_tokens_repriced": true, "input_usd_per_million": 4, "output_usd_per_million": 12, "cached_read_usd_per_million": 1 }, "us_regional_endpoint_multiplier": 1.1, "selected_endpoint": "global", "promotional_period": null }, "status": "verified_current_official_cached_read_tariff", "sourceURL": "https://docs.x.ai/developers/pricing", "sources": [ "https://docs.x.ai/developers/pricing", "https://docs.x.ai/developers/advanced-api-usage/prompt-caching/usage-and-pricing" ], "notes": [ "Official pricing table lists $0.50 short-context and $1 long-context cached input.", "First request or eviction is a cache miss with full prompt-token pricing. No separate write/storage tariff is listed; null is not a zero-price claim.", "Pricing table uses >=200k; cache guide says exceed. Reference requests remain well below the boundary." ], "checkedAt": "2026-10-08T03:22:13Z" } }, "benchmark_records": [ { "benchmark": "GPQA Diamond", "mean_score": 0.92739899, "mean_score_display": "0.92739899", "standard_error_percentage_points": 1.551, "se_display": "1.551", "started_at": "2026-09-22T21:49:22.000Z", "record_id": "hxAhYJfXdPNanxatXpG5hC", "csv": "gpqa_diamond.csv", "protocol_url": "https://epoch.ai/benchmarks/gpqa-diamond" }, { "benchmark": "FrontierMath v2", "mean_score": 0.52982456, "mean_score_display": "0.52982456", "standard_error_percentage_points": 2.962, "se_display": "2.962", "started_at": "2026-09-22T21:49:22.000Z", "record_id": "F4unBbW5M4VJogzCq4vGAb", "csv": "frontiermath_tiers_1_3_v2.csv", "protocol_url": "https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2" } ] } ] }