LLM API pricing

Full pricing table

Conditions attached to headline prices

30 of the models we track carry a condition that changes what you actually pay. These matter more than a few cents per million tokens.

• GPT-5.6 Sol — Promotional pricing (down from $5/$30) held at least through 2026-11-21. Prompts above 272K input tokens bill at higher rates; 'fast' processing is 2x.
• GPT-5.6 Terra — Cut 20% from the $2.50/$15 launch price on 2026-07-30.
• GPT-5.6 Luna — Cut 80% from the $1/$6 launch price on 2026-07-30.
• Claude Fable 5 — 5-minute cache writes $12.50/MTok, 1-hour $20/MTok. Requires 30-day data retention, so not available under zero-data-retention terms.
• Claude Mythos 5 — Same price as Fable 5. Access restricted to Project Glasswing partners.
• Gemini 3.1 Pro — Prompts over 200K tokens reprice to $4 input / $18 output per 1M.
• Gemini 3.5 Flash — Flat rate at any context length, unlike Gemini 3.1 Pro's 200K cliff.
• DeepSeek V4-Pro — Off-peak rate shown. Peak rates are exactly double ($1.32 input / $3.96 output) during 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday. Cache hits cost roughly 3% of a cache miss.
• DeepSeek V4-Flash — Off-peak rate shown. Peak rates double ($0.44 / $1.32) during 01:00–04:00 and 06:00–10:00 UTC on weekdays. Superseded the earlier flat-rate pricing of $0.14 / $0.28 that many comparison sites still quote.
• GLM-5.2 — Independently measured by Artificial Analysis; the vendor's own pricing page was unreachable at time of checking. Treat as approximate and confirm in the console.
• Kimi K3 — The custom licence requires a separate commercial agreement once the licensee and its affiliates exceed $20M revenue over any consecutive 12 months.
• Universal-2 — Batch transcription. Speech-understanding add-ons are billed separately.
• GPT-4o Mini Transcribe — The OpenAI transcription endpoint caps uploads at 25 MB, which a single long recording will exceed.
• Scribe v2 — Published as $0.22 per audio hour. Entity detection adds $0.07/hour and keyterm prompting $0.05/hour. A realtime variant runs $0.39/hour (~$0.0065/min) at roughly 150 ms latency.
• Nova-3 — Batch pay-as-you-go rate for the monolingual model. Streaming is higher and quoted between $0.0048 and $0.0077/min depending on plan and source; multilingual adds roughly 20%. Confirm your tier in the console.
• GPT-4o Transcribe — Legacy whisper-1 costs the same. Live transcript streaming through the realtime endpoint is far pricier at roughly $0.017/min. Uploads are capped at 25 MB.
• Flash / Turbo TTS — Published as $0.05 per 1,000 characters. Text-to-speech bills per character, not per audio minute.
• Multilingual v3 TTS — Published as $0.10 per 1,000 characters. Voice changer and isolator are $0.12/min, sound effects $0.12/min, dubbing $0.33–$0.50 per source minute.
• Imagen 4 Fast — 1K resolution.
• FLUX.2 [pro] — 1024² generation. The [pro] tier is API-only; FLUX's smaller schnell and dev variants are the openly downloadable ones.
• GPT Image 1.5 — Medium quality at 1024px. High quality is reported between $0.13 and $0.17 per image and sources disagree on the exact figure — confirm in the console before budgeting. A cheaper Mini variant runs about $0.036 at high quality.
• Nano Banana 2 — 1024² render. Priced at $60 per million output tokens, so cost tracks resolution: $0.045 at 512px, $0.067 at 1K, $0.101 at 2K, $0.151 at 4K. Batch mode halves all of these.
• Nano Banana Pro — 1K–2K resolution on the standard tier; 4K is $0.24. Batch and Flex lanes cut both roughly in half, to $0.067 and $0.12.
• Veo 3.1 Lite — 720p, video only, no audio. Quoted between $0.03 and $0.05/sec depending on route. Clips are 4, 6, or 8 seconds; access is quota-limited and regional.
• Runway Gen-4 Turbo — Billed as 5 credits per second at $0.01 per credit. No native audio on the Gen-4 family.
• Kling 3.0 — 720p without native audio. Adding audio takes it to roughly $0.11–$0.13/sec, and 1080p to $0.14–$0.168/sec. Clips run 3–15 seconds.
• Runway Gen-4.5 — Billed as 12 credits per second at $0.01 per credit. No native audio on the Gen-4 family.
• Veo 3.1 Fast — Audio included. Video-only routes are cheaper at roughly $0.08–$0.10/sec, and third-party resellers vary. Clips are 4, 6, or 8 seconds.
• Veo 3.1 — 720p/1080p with audio included. Video-only routes run about $0.20/sec and 4K is priced separately. Clips are 4, 6, or 8 seconds; maximum 4 outputs per request.
• Sora 2 — The API shuts down on 24 September 2026 — do not start new work on it. 720p standard; the Pro tier ran $0.30/sec at 720p and $0.70/sec at 1080p. Batch was half price. Veo 3.1 Fast and Kling 3.0 are the closest replacements.

Frequently asked