Head-to-head showdown: GPT-5.3 Codex ($1.75 in / $14.00 out per 1M) vs AI21 Jamba 1.5 Large ($2.00 in / $8.00 out per 1M). AI21 Jamba 1.5 Large is 1.6× cheaper across standard token mixes, with 256K vs 256K context windows.
gpt-5.3-codex · openai
jamba-1.5-large · ai21
Benchmark
| Workload Scenario | GPT-5.3 Codex | AI21 Jamba 1.5 Large | Price Delta |
|---|---|---|---|
| 1M input tokens (raw text) | $1.75 | $2.00 | −$0.25 |
| 1M output tokens (generation) | $14.00 | $8.00 | +$6.00 |
| 1M tokens · 70% input / 30% output mix | $5.425 | $3.80 | +$1.625 |
| Standard chat turn (4K in / 800 out, 50% cached) | $0.0151 | $0.0144 | +$0.00065 |
| Monthly scale (10K requests / day) | $4,515.00 | $4,320.00 | +$195.00 |
Negative difference = GPT-5.3 Codex is cheaper. Positive = AI21 Jamba 1.5 Large is cheaper.
Scaling Curve
Total cost of a token volume at 70/30 input/output split (uncached). Log-log scale.
Verdict
GPT-5.3 Codex has the cheaper input rate, while AI21 Jamba 1.5 Large has the cheaper output rate. The crossover happens when output makes up about 4% of your total tokens.
Below that share (retrieval, summarization, extraction — lots of context in, little text out) GPT-5.3 Codex is cheaper. Above it (generation, translation, coding — long completions) AI21 Jamba 1.5 Large wins.
FAQ
On input, GPT-5.3 Codex is cheaper ($1.75/M vs $2.00/M). On output, AI21 Jamba 1.5 Large is cheaper ($8.00/M vs $14.00/M). For workloads where more than 4% of tokens are output, the output-cheaper model wins overall.
A chat-style request (4,000 input + 800 output tokens, 50% cached) costs $0.0151 on GPT-5.3 Codex and $0.0144 on AI21 Jamba 1.5 Large — AI21 Jamba 1.5 Large is 1.0× more expensive for that workload.
GPT-5.3 Codex supports 256,000 tokens (64K max output); AI21 Jamba 1.5 Large supports 256,000 (4.1K max output). They are equal.