{"task": "coding", "filters": {"budget_per_1m": null, "providers": "all", "features": "any", "input_modalities": "any", "output_modalities": "any", "service_tier": "standard", "context_tokens": null}, "ranked": [{"rank": 1, "id": "claude-opus-5-5", "model_id": "claude-opus-5-5", "provider": "anthropic", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $4.0/1M input; strong in: reasoning; 1M context \u2014 Anthropic's recommended default (2026-09-21): 'start with Claude Opus 5.5 for most workloads'. Same $4/$20 as Opus 5 but", "context_window": 1000000, "modality": {"input": ["text", "image", "pdf"], "output": ["text"], "endpoints": null, "max_output_tokens": 128000, "source": "models.dev"}, "cost": {"input_per_1m": 4.0, "output_per_1m": 20.0, "cached_input_per_1m": 0.2, "cache_write_per_1m": 5.0}, "cost_ladder": {"service_tiers": {"batch": {"input_per_1m": 2.0, "output_per_1m": 10.0}}, "context_step": null, "reasoning_output_per_1m": null, "effort_supported": ["xhigh", "max"], "source": "litellm", "supports_reasoning": true}, "effective_input_per_1m": 4.0, "aedelgard_note": "Anthropic's recommended default (2026-09-21): 'start with Claude Opus 5.5 for most workloads'. Same $4/$20 as Opus 5 but a Jun-2026 cutoff and cache reads at 0.05x ($0.20 vs $0.50), the cheapest flagship per cached token in the catalogue. Fable 5.1 remains the ceiling for long-horizon agentic work. 5m cache write $5 / 1h $8. [read 2026-09-23]"}, {"rank": 2, "id": "nvidia/Nemotron-3-Ultra-550b-a55b", "model_id": "nvidia/Nemotron-3-Ultra-550b-a55b", "provider": "nebius", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $1.0/1M input; strong in: reasoning \u2014 Auto-classified (2026-09-25): tier=flagship from family/size; price from litellm:nebius/nvidia/Nemotron-3-Ultra-550b-a55", "context_window": 128000, "modality": {"input": ["text"], "output": ["text"], "endpoints": null, "max_output_tokens": 1048576, "source": "models.dev"}, "cost": {"input_per_1m": 1.0, "output_per_1m": 3.0, "cached_input_per_1m": null, "cache_write_per_1m": null}, "cost_ladder": null, "effective_input_per_1m": 1.0, "aedelgard_note": "Auto-classified (2026-09-25): tier=flagship from family/size; price from litellm:nebius/nvidia/Nemotron-3-Ultra-550b-a55b. Editorial review welcome."}, {"rank": 3, "id": "moonshotai/Kimi-K3", "model_id": "moonshotai/Kimi-K3", "provider": "nebius", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $3.0/1M input; strong in: reasoning \u2014 Auto-classified (2026-09-25): tier=flagship from family/size; price from litellm:nebius/moonshotai/Kimi-K3. Editorial re", "context_window": 131072, "modality": {"input": ["text"], "output": ["text"], "endpoints": null, "max_output_tokens": 1024000, "source": "models.dev"}, "cost": {"input_per_1m": 3.0, "output_per_1m": 15.0, "cached_input_per_1m": null, "cache_write_per_1m": null}, "cost_ladder": null, "effective_input_per_1m": 3.0, "aedelgard_note": "Auto-classified (2026-09-25): tier=flagship from family/size; price from litellm:nebius/moonshotai/Kimi-K3. Editorial review welcome."}, {"rank": 4, "id": "deepseek-ai/DeepSeek-V4-Pro-0813", "model_id": "deepseek-ai/DeepSeek-V4-Pro-0813", "provider": "nebius", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $0.55/1M input; strong in: reasoning; 1M context \u2014 DeepSeek 862B MoE (checkpoint 0813) hosted in Finland via Nebius Token Factory.", "context_window": 1000000, "modality": {"input": ["text"], "output": ["text"], "endpoints": null, "max_output_tokens": 979000, "source": "models.dev"}, "cost": {"input_per_1m": 0.55, "output_per_1m": 2.19, "cached_input_per_1m": 0.14, "cache_write_per_1m": 0.55}, "cost_ladder": null, "effective_input_per_1m": 0.55, "aedelgard_note": "DeepSeek 862B MoE (checkpoint 0813) hosted in Finland via Nebius Token Factory."}, {"rank": 5, "id": "deepseek-ai/DeepSeek-V4-Pro", "model_id": "deepseek-ai/DeepSeek-V4-Pro", "provider": "nebius", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $0.55/1M input; strong in: reasoning; 1M context \u2014 DeepSeek 862B MoE flagship hosted in M\u00e4nts\u00e4l\u00e4, Finland via Nebius Token Factory. Direct EU jurisdiction.", "context_window": 1000000, "modality": {"input": ["text"], "output": ["text"], "endpoints": null, "max_output_tokens": 1048576, "source": "models.dev"}, "cost": {"input_per_1m": 0.55, "output_per_1m": 2.19, "cached_input_per_1m": 0.14, "cache_write_per_1m": 0.55}, "cost_ladder": null, "effective_input_per_1m": 0.55, "aedelgard_note": "DeepSeek 862B MoE flagship hosted in M\u00e4nts\u00e4l\u00e4, Finland via Nebius Token Factory. Direct EU jurisdiction."}, {"rank": 6, "id": "NousResearch/Hermes-4-405B", "model_id": "NousResearch/Hermes-4-405B", "provider": "nebius", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $1.0/1M input; strong in: reasoning \u2014 Auto-classified (2026-09-25): tier=flagship from family/size; price from litellm:nebius/NousResearch/Hermes-4-405B. Edit", "context_window": 128000, "modality": {"input": ["text"], "output": ["text"], "endpoints": null, "max_output_tokens": 131072, "source": "models.dev"}, "cost": {"input_per_1m": 1.0, "output_per_1m": 3.0, "cached_input_per_1m": null, "cache_write_per_1m": null}, "cost_ladder": null, "effective_input_per_1m": 1.0, "aedelgard_note": "Auto-classified (2026-09-25): tier=flagship from family/size; price from litellm:nebius/NousResearch/Hermes-4-405B. Editorial review welcome."}, {"rank": 7, "id": "gpt-6-sol", "model_id": "gpt-6-sol", "provider": "openai", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $2.0/1M input; strong in: reasoning; 1M context \u2014 GPT-6 workhorse flagship (2026-09-14), built for complex coding and agentic workflows. $2/$10, half of GPT-5.6 Sol and l", "context_window": 1050000, "modality": {"input": ["text", "image", "pdf"], "output": ["text"], "endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], "max_output_tokens": 128000, "source": "litellm+models.dev"}, "cost": {"input_per_1m": 2.0, "output_per_1m": 10.0, "cached_input_per_1m": 0.2, "cache_write_per_1m": 2.5}, "cost_ladder": {"service_tiers": {"batch": {"input_per_1m": 1.0, "output_per_1m": 5.0}, "priority": {"input_per_1m": 4.0, "output_per_1m": 20.0}}, "context_step": {"above_tokens": 272000, "input_per_1m": 4.0, "output_per_1m": 15.0}, "reasoning_output_per_1m": null, "effort_supported": ["none", "xhigh", "max"], "source": "litellm", "supports_reasoning": true}, "effective_input_per_1m": 2.0, "aedelgard_note": "GPT-6 workhorse flagship (2026-09-14), built for complex coding and agentic workflows. $2/$10, half of GPT-5.6 Sol and level with Claude Sonnet 5, at a 1.05M window (922k max input) / 128k output, cutoff Apr 20 2026. Effort none/low/medium(default)/high/xhigh/max. EU data residency (Standard processing only). OpenAI's better default than Astra on cost. [read 2026-09-23]"}, {"rank": 8, "id": "gpt-6-astra", "model_id": "gpt-6-astra", "provider": "openai", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $10.0/1M input; strong in: reasoning; 1M context \u2014 OpenAI's top model (2026-09-05): 'most capable model, built for the hardest end-to-end work'. 1.05M window (922k max inp", "context_window": 1050000, "modality": {"input": ["text", "image", "pdf"], "output": ["text"], "endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], "max_output_tokens": 128000, "source": "litellm+models.dev"}, "cost": {"input_per_1m": 10.0, "output_per_1m": 50.0, "cached_input_per_1m": 1.0, "cache_write_per_1m": 12.5}, "cost_ladder": {"service_tiers": {"batch": {"input_per_1m": 5.0, "output_per_1m": 25.0}, "priority": {"input_per_1m": 20.0, "output_per_1m": 100.0}}, "context_step": {"above_tokens": 272000, "input_per_1m": 20.0, "output_per_1m": 75.0}, "reasoning_output_per_1m": null, "effort_supported": ["xhigh", "max"], "source": "litellm", "supports_reasoning": true}, "effective_input_per_1m": 10.0, "aedelgard_note": "OpenAI's top model (2026-09-05): 'most capable model, built for the hardest end-to-end work'. 1.05M window (922k max input), 128k output, cutoff Apr 30 2026, effort low->max. Price peer of Claude Fable 5.1 ($10/$50): the ceiling, not the default; gpt-6-sol at $2/$10 covers most days. [read 2026-09-23]"}, {"rank": 9, "id": "claude-fable-5-1", "model_id": "claude-fable-5-1", "provider": "anthropic", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $10.0/1M input; strong in: reasoning; 1M context \u2014 Fable-class for long-horizon agents. Same $10/$50 as Fable 5 but cache reads are 4\u00d7 cheaper ($0.25 vs $1.00), so it is t", "context_window": 1000000, "modality": {"input": ["text", "image", "pdf"], "output": ["text"], "endpoints": null, "max_output_tokens": 128000, "source": "models.dev"}, "cost": {"input_per_1m": 10.0, "output_per_1m": 50.0, "cached_input_per_1m": 0.25, "cache_write_per_1m": 12.5}, "cost_ladder": {"service_tiers": {"batch": {"input_per_1m": 5.0, "output_per_1m": 25.0}}, "context_step": null, "reasoning_output_per_1m": null, "effort_supported": ["xhigh", "max"], "source": "litellm", "supports_reasoning": true}, "effective_input_per_1m": 10.0, "aedelgard_note": "Fable-class for long-horizon agents. Same $10/$50 as Fable 5 but cache reads are 4\u00d7 cheaper ($0.25 vs $1.00), so it is the better Fable for cache-heavy persistent agents; supersedes Fable 5. 5\u00d7 Sonnet 5's price: use only when the work needs Fable-tier judgement. [anthropic.com/pricing 2026-09-04]"}, {"rank": 10, "id": "gemini-3.1-pro-preview", "model_id": "gemini-3.1-pro-preview", "provider": "google", "tier": "flagship", "score": 10.0, "reason": "flagship tier; $2.0/1M input; strong in: reasoning; 1M context \u2014 Google's 3rd-gen Pro, still labelled preview. $2/$12 (\u2264200k prompts; $4/$18 above). Prefer over gemini-2.5-pro for new w", "context_window": 1048576, "modality": {"input": ["text", "image", "video", "audio", "pdf"], "output": ["text"], "endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], "max_output_tokens": 65536, "source": "litellm+models.dev"}, "cost": {"input_per_1m": 2.0, "output_per_1m": 12.0, "cached_input_per_1m": 0.2, "cache_write_per_1m": null}, "cost_ladder": {"service_tiers": {"batch": {"input_per_1m": 1.0, "output_per_1m": 6.0}, "priority": {"input_per_1m": 3.6, "output_per_1m": 21.6}}, "context_step": {"above_tokens": 200000, "input_per_1m": 4.0, "output_per_1m": 18.0}, "reasoning_output_per_1m": null, "effort_supported": null, "source": "litellm", "supports_reasoning": true}, "effective_input_per_1m": 2.0, "aedelgard_note": "Google's 3rd-gen Pro, still labelled preview. $2/$12 (\u2264200k prompts; $4/$18 above). Prefer over gemini-2.5-pro for new work; watch for the GA id. [ai.google.dev pricing 2026-09-04]"}], "candidate_pool": 113, "updated_at": "2026-10-12T06:00:48Z", "ttl_seconds": 3600}